#📚┃suno-school
1 messages · Page 9 of 1
Where do I put outro
In the lyrics field. you can also use [intro], [chorus], [verse], in lyrics
[Outro]
So, my love, hold on tight,
Together we'll make it, through the darkest night,
I'll always love you, through thick and thin,
In your rough times, our love will always win.
Like that?
Put [outro] after the lyrics, and use less lines, 4 might be too much. Even 0 lyrics is best sometimes, just [outro] alone. However sometimes outro in front can work -- not a strict rule.
But for chorus and verse, use like that
So, my love, hold on tight,
Together we'll make it, through the darkest night,
I'll always love you, through thick and thin,
In your rough times, our love will always win. [Outro]
Like that?
Close, put it on a line by itself, and also add an empty line after the lyrics. And if it doesn't work, then try using only 2 lines of lyrics intead of 4, then try with 0 lines of lyrics. You can also try both, for example
So, my love, hold on tight,
Together we'll make it, through the darkest night,
I'll always love you, through thick and thin,
In your rough times, our love will always win.
[Outro]
[fade out]
Overall getting endings is tricky. If you search discord for [outro] you can find examples
How do I do that
I cut it down to two lines and then gonna do the last two lines
just search the word 'outro' in desktop discord, search on top right. not sure on mobile but should be somewhere. many chirp samples are in the public chirp rooms so you can look at what other people did and see what works or doesn't work, and get ideas
And what did you mean by try with zero lines of lyrics? What would happen? Just a closing “song” sorry of?
The idea with no lyrics is the song has no more singing, just the music continues to an ending
Oh! Nice! Thank you!
But no matter what you have to be a little lucky to get a good ending. At least in this version. Endings (and intros) should be better in later versions though
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
How do you write a song longer than 40 seconds
hello, is it possible to keep same music & same rythm than an existing song ?
and what are the existing styles ? because if I put "80's" he didn't find
Continue button, then heart on last clip, see #⏩┃continuation message
80s is probably too general, but 80s + other standard music like pop, rock, metal, etc, should be okay. No fixed list but generally the main genres should work better, and you can load up the style prompt with both general and specific ones to cover your bases and hedge a bit.
maybe something like that in a future version but not now
actually 80s doesn't work that badly if you just stack up a bunch of 80s. not the greatest music though, more descriptive style prompt would be better
Reminds me of the ones on Bojack Horseman for time periods:
These are the generic pop songs of the 80s, 90s, and 2007.
80s/90s is from Season 1 Episode 8.
2007 is from Season 3 Episode 2.
Like I said, I'm not accusing anyone of anything. I was just frustrated at the moment. I like what I get a lot of the time with Chirp, it's just so frustrating to have gotten good results and then be later staring at the thing and wondering how the heck it worked the last time and why it's not working now.
I'm trying to learn to see each song as a journey to somewhere new.
So what exactly does it mean when you use tags and they combine?
(Sharing the chirp because even though I don't want to use it and there were no vocals, that's an awesome sax...)
That's actually just for the video to make more stuff fit. under the hood the model is respecting the text you put (newlines etc)
So it's not actually a positive or a negative, it's just a formatting for space thing?
I got the idea in my head that sometimes it meant that the tags that combined were official but undocumented tags. 😏
Sorry, you start siezing on every little thing after a while to try to figure this stuff out. 😅
How do you make it's male or female voice?
try 'male vocals' in the description
doesn't always work but helps
my credit are over now wht to do ?
yo @versed valley said he flagged my tune and wont he wont give it me.
try different lyrics and try keeping it as pg as possible for the filters to not flag it
can someone tell me how to get the dialog box to come up to generate a song. When i type /chirp in the message bot , nothing happens.
I found instead of saying generic music genres. You're much better off saying the instruments and type. Chat GPT is pretty good at giving you prompts for detailing music genres in better broken down prompt ways.
you get 25 free chirps every month!
But, it's only ten dollars a month for 100 more chirps a month!
I think the most common issue done here is you aren't typing /chirp in the right channel. Go to #🐣┃chirp-beta-3 or DM /chirp to the @versed valley directly.
Found the issue. I wasn't hitting enter twice. when I hit it twice, it came right up. thank you for responding.
Glad you got it figured out 🙂
We can’t say KISS? -_-
I could before...I think...
damn... i'm stupid... well, mostly new here, hehe... i can't seem to grasp the lower half of "add /chirp bot to server" image you posted. where do i get this black panel with "add to server" button ? 🤪
You can put both in the prompt together too, genre, instruments, and even general descriptions like sad or happy. using chatgpt to get a nice detailed description is a good workflow for sure.
You click on the bot itself. It may look different on mobile. If you have a Discord server looks like this. (Just setup a test to confirm right now)
Which do your use / think works better:
[Male Vocal]
[Male Voice]
[Male Singer]
what about brute force Male, Male, Male, Male, Male, Male, Male, Male, Male - though it may have side effects on the music output... but 8/8 singer gender match testing just now
🤔 that 2nd clip is a male voice?
second clip is Female, Female, etc.. did 4 of each and all 8 clips matched.
OH! (der) i though the visual was the same, and didn't even notice it was 2 different clips 😓
(but actually I was asking about Voice, Vocal, or Singer....
(but yeah, gender is a very common question)
But brute forcing anything ... hmmm. I will try next time Chirpbot is stubborn
Yeah, AI can be like that.
ha! now, now... thx! 😍
Chirp is surprisingly bad in creating "chillhouse" generations: the voice is always terrible
I tried 3 times...
Have you compared it to other styles? Right now I'm getting a horrible hiss in my voices and I don't know how to fix it.
Wondering if we hit a bug?
...
Some of the music was really on point too, so I get why you got frustrated.
....
Changing the subject, how do we feel about cover songs?
This is the first time I've done one.
Basically after beating my head against the wall I started playing with good lyrics rather than my doggeral to see if I could at least get a song today. So far everything I've done has been crud today.
Meh...That hiss is back. 😦
I'm pitching the effort.
But for future reference?
Giving some lyrics from ChatGPT a try. At least I know there's no entanglements there.
/Chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
...damn,,,,I'm going to need a new virtual group again, aren't I?
My friend sent me a song made with chirp and I wanted to continue it myself, but they generated it in DMs. Is there a way for me to continue it by, like, getting the seed or something?
Not in current version. Unless you can add somebody else to a DM and make it a group DM? I'm actually not sure if that is a feature in Discord. But in future you can add Chirp Bot to your own Discord so you can both see the private gens and continue each other's outputs
There's no maximum limit but it's often hard to stay consistent for more than 2 or 3 minutes, and sometimes audio quality has issues the longer you go. Not always there's a 17 minute song somebody posted today, but often people have trouble going longer than that and staying high quality.
I got lucky and generated a really good song that everything after it has been bad
So damn
Also what happens if I try to continue a song that hasn't finished singing the lyrics, like it got cut midway? Can I make it finish it?
You can try, but if it cut off at a really awkward place, like right in the middle of the word, it's usually very hard to get a smooth connection. But if the break is a little more natural, at a place where the singer pauses anyway, it can work.
Maybe sooner than that...
If the song cuts off at end of a line of lyrics, then usually it works well to finish it
Tbh making it more accessible would be better, like a more efficient algorithm that would allow free tier people to make more songs a month
A lot of it boils down to regenerating a single song until you get it good so it takes a lot of uses
Yeah. One way to be efficient with credits is to browse the chirp public channels, and start your song as a continue instead, picking somebody else's sample you already really like.
Alright. I guess I could cut off the rest after I finish it
Sorry, I had to leave the house after I wrote that message. I will try now, a bug is totally possible
I always give lyrics from GPT, I like creating something completely new
I actually have some idea of fine-tuning GPT-3.5 with a bunch of song lyrics, and se what comes out. I figured out even GPT-4 is actually very bad in writing lyrics, they always come out more like perfectly rhyming nursery rhymes, instead of actual songs, too perfect and meaningless
and it's very difficult to make it NOT rhyme or to make some less-than-perfect rhymes. I want to see if fine tuning gives me better results
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
I've been writing my own lyrics mostly, up to now. Oh I've been ChatGPT inspired don't get me wrong. I've asked for lyrics from it. But I didn't leave them the same. I sat there and rewrote them and tweaked them until they were mine up to this most recent one.
Even then, I don't usually follow the pattern it gives when seeking inspiration. I'll take those six verses and do them out of order, I'll rewrite so that instead of six short verses I have two or three longer ones. Stuff like that.
This is another thing--how many rhymes is too many rhymes? I hear this a lot and it stumps me a bit because I actively try to put more rhymes in when I can to give things a better rhythmic flow.
It's funny, every once in a while I'll find myself looking at lyrics and being shocked at how little rhyming there is. Always seems like there was more when I remember it.
My biggest trouble seems to be the choruses. I can generally develop some lyrics and string them together. Getting the tones right for a chorus and returning to the voice in the lyrics and being able to continue from there without the whole thing de-evolving on me is hard. All too often it seems like what I thought was a great chorus just falls apart after listening.
Hey there, is there a way to effectively force the system to have a woman's voice ?
I'm over 10 tries with ONLY male voices
Tried "girl voice" at the end of the style prompt, at the beginning of the lyrics prompt I tried : [Girl] or [Girl Voice] but nothing seems to change something
see this advice: #📚┃suno-school message
Is it possible to have the AI have the style of a specific singer or song?
how i see my credits to use Chirp?
Is accapella a viable style? I want to avoid the backing tracks. But I also want to sing in a specific genre typically. Do I have to separate the vocals with 3rd party?
The model doesn’t recognize specific artists, but you can describe their styles in the first field in /chirp command's pop-up: Describe the Type of Music field
You can use the /info command, which doesn't appear to work in this channel at the moment. We'll fix this.
In the meantime, you can use the /info command in other channels like #💬┃general-chat or view account details — including available credits — at https://app.suno.ai/account
Yes, albeit with mixed results for the time being. Try typing 'accapella' or 'vocals only' in the Describe the Type of Music field! You can also search for accapella in the public generation channels like #🐣┃chirp-beta-1, #🐣┃chirp-beta-2, etc. for inspiration
Yeah that's worked a treat. Have an amazing day very helpful nice person
is there a way to use same seed as previous chirp, as in try to make generally identical chirp but make changes to lyrics?
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Has the quality really gone to shit in the last 12 hours? everything i'm generating sounds like poor quality today
How do I specify the language in which I want the voice to sing in?
That's a good point, about picking some lyrics and developing from there. My problem is "ain't nobody got time for that". Plus my brain is naturally attracted by the music way more than the meaning. So I basically never pay attention to the lyrics, only how they flow in the song. But of course I admire whoever can write compelling lyrics...
About this, sometimes I have luck by saying "Remove all the rhymes from the song". Due to the randomness of GPT, it doesn't work all the times, but sometimes it does, and it leaves just enough rhymes, way less perfect than before but more interesting
About this: I generate a few parts and I assemble them with BandLab Studio. It has a beat counter, so you can easily and precisely cut your generations at the right point and edit and mix them
So you basically generate the first part, then generate (hopefully!) an interesting chorus, then generate a second part directly as a continuation of the first one, then edit everything together
Check here, these are the best things I generated. There's mostly editing and stereo effect, no other effects whatsoever. Only for one song (lost in space) I used a noise reduction effect and more echo.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
I admit to having given some consideration to taking parts, stripping the music to get just the vocals and applying effects. Lost in Space is making me consider my consideration, I love those vocal effects!
When I can get Chirp to do it for me, I'm always estatic. Being able to just apply them myself would be a huge bonus.
up: is there a way to use same seed as previous chirp, as in try to make generally identical chirp but make changes to lyrics?
Theoretically, wouldn't this be possible by just reusing a starting point after completing a song?
You'd have to be careful with how you worded stuff from the beginning to avoid drift and there is always a bit of a disce throw for it but theoretically I assume it should be possible as is.
Do you mean that if I just continue the song with changed lyrics from the first part onwards? My problem is that I want to change words precisely from the first part, so I guess this idea doesn't work in this case.
This is a feature that's been discussed as a possibility for future versions, but not in current version. But eventually they want to get something like this as a feature. Sometimes if you are very lucky then a regular song continue can be very consistent and almost work like this, but it's a roll of the dice and have to get fairly lucky. You can try tweaks like using small song chunks as the main part you continue from, less than 30 or 40 seconds, that may make it a bit more likely.
You don't have to specify a language, you just use the language and it should sing in the right language.
Hello team, some of the Chirp is taking longer than usual or not generating at all. Just a heads-up.
I am seeing that also.
Can anyone give me some feedback on joining parts? I have a great chorus and I managed to go backwards for another set of verses but I can't seem to get that chorus to come in naturally when I try to join it. It's driving me crazy.
sometimes (Chorus) works better for me that [Chorus] - I can't be wrong but yeah lol
I guess I can go back again and try it.
I've done everything else.
Hate when stuff starts coming undone half way like this, especially when it seemed to be going so well to begin with.
Maybe you could try continuing the segment you want to join with already generated chorus, but try to make a maximally 'connective' clip, maybe a very short, with a bit of break or pause in the music at the very end. I haven't tried this in Chirp so not sure it will make it join better in post, but something similar to that has been useful in Bark for me. Edit: Oh I may have misunderstood what you were trying to do. I thought you were joining the already existing chorus audio to later clips yourself, manually. So that's what I'm describing, if it's not clear.
Yeah, I am...here's the best that my meager skills could do in that regard. First join is at about 2:48 seconds in. Then the chorus comes in abou 3:45ish...
It's a noisy join though because the chorus is noisy. 😦
I accidentially a letter.
Ha! I noticed a spot that I missed before in the waveform that showed a good place to splice. I think I actually got it now!
hi I'm just asking if you have a solution for making an angry rap? a bit like scarlxrd? or some xxxtentacion scream rap, What style of sound should you use? THANKS
Maybe things like Rage Rap, Scream Rap, Angry Rap, Trap, Rage, Angry, Rage, Rage, just tossed them all in so not sure what's actually working. (not too familiar with the music style actually so this may not be correct, but it IS angry, at least) #🐣┃chirp-beta-7 message
thks bro !
I tried one with just a single two word prompt and it was rap but not angry, that's why I went overboard on that one and stacked up the Rage words. Probably better somewhere in between...
rage rap is working for me 😄 thks
i mix, rage rap and alternative rock to have what i want but is cool ! 😁
Well, I guess in a few months from now we will be able to control the generation much more than now. I see the need of editing as a fun inconvenience, rather than a feature. I'm a lazy person, I know...
I'm extremely sympathetic to that point of view. I consider it cheating. But sometimes its the only way that works.
I feel this so much. I get overly obsessed with the idea of not cheating in every AI thing not just audio - just a real 100% genuine AI creation. And then tunnel vision on it and waste 10s of hours beating my head against a wall... lol. (Most recently trying to get Bark to compose a full song lyrics that make sense, which frankly I haven't given up on... it's too funny of an idea not to be done.)
Exactly. After a bit you have to ask yourself if you're letting the perfect be the enemy of the good.
That and my other obsession, a love of using "the wrong tool for the job" and it's a dangerous combo for destroying time. At one point I tried using Chirp to generate new Magic the Gathering cards, via unconditional lyrics-free continues. As you might expect it doesn't work... yet. Maybe in a bigger and better future Chirp... (It works a tiny tiny bit in Bark, but you can't get a legal playable deck or anything out of it and most 'new' card creations are extremely simply word swaps that mostly just repeat a line from the prompt.)
It was kind of funny though...
The fact it even came close...
It should possible for real, some day... The better it gets at continuing audio, well, at some point the way to get better is to start spitting out lyrics with real cards... a momentous occasion for all those people (ie, me) waiting for the day they can play MTG decks generated by an AI music model.
@untold briar sorry for tagging you, idk if you know about this, i've a question :
Are there any special words (for style of song) ? words that will really impact the sound? Does the order matter?
I saw that there is already a big difference when we mark intro versus chorus bridge etc. but are there any hidden words? like easter eggs or something like that? thks
or maybe the Suno team idk
This is a good question.
For an instance, does it matter whether or not [Instrumental] comes before or after the instruments named?
Order doesn't matter, for words separated by commas. (It matters for compound words though, like rock-pop would be different than pop-rock, maybe. That's just one word.) Order 100% matters in the LYRICS field for anything though.
The answer is basically yes to Easter eggs, sort of. The words aren't hidden by the Suno team on purpose. It's just that the AI is trained, the AI sees a lot of things, and it learns a lot of stuff. Some of that stuff is put there on purpose by Suno devs, yes, but the AI also leans tosns of words and how to do things that nobody really put there on purpose -- it was just in the training data. So even Suno has to discover all the best words to use for Bark and Chirp by testing.
Okay thank you that can make a cool mix ^^ but I especially wanted to ask if there are words that are stronger than others, or special ones, is there a list of "secret" chirp words? x)
Some words are 100% stronger than others. One thing you can do is repeat words. Like Country, EDM, EDM, EDM, if you want just a little country, but more EDM
Country in particular seems STRONG and overpowering
So it's possible to put the name of a singer and with a little luck that can give us something similar?
Singer names specifically aren't in the music style field, though you can put some pretty strange things in there and it sometimes works a bit.
mmh ok thanks bro ! 💜
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
@untold briar I'm trying to extend my song. I did two generations fine, but the third one keeps babbling random words. Am I doing something wrong?
/CHIRP
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
is there a way to make chirp end a song with proper ending, not just cut abruptly?
What is the 'proper ending' that you want? I have used all of the below (not at the same time, and in a Continuation ⏩ clip with no other lyrics)
[End]
[Fade Out]
[Outro]
[Refrain]
(Refrain has lyrics)
Some work 'in-bot', and some repeat the music so you can fade out in an audio editor.
And being AI they are not 100% reliable.
thanks, will try those!
I ran out of "Chirps" and would like to buy more, but when I try to access the payment screen, I am informed by Discord, "You need to verify your account in order to perform this action." I've had this account for a long time and I am already verified. (Just to make sure, I looked up instructions on line.) How can I pay for additional Chirps? I right in the middle of a hot #1 song for halloween.
hm sry let me ask @light ether
I figured out it was because I didn't have a phone number linked to the account. It didn't tell me what the issue was, I just tried adding my phone number and it worked.
OK, technical problem with my song production: I used Verse-Chorus-Verse-Chorus-Bridge...
Now I need one more chorus and an outro. But the last Choruses I'm generating are starting to go off the rails. Is there any fix for this?
no. the next clip can only look back at 10sec from the 'bridge' clip.
you might be able to use an audio editor to stitch various takes together (kinda depends on the audio).
you could go back to a Continue point that gave you a good chorus, and regenerate your 3rd chorus from there. It will still be different (because AI) but the sound quality and beats will be from the earlier part of the song.
OK, that's a good idea, thanX.
What exactly do you mean about looking back 10s? Is that from specifically the bridge, or does it have to do with the number of generations, or???
You can also get some good results with the tools here:
https://vocalremover.org/
The context window is about ten seconds long. So to get the continuity the AI looks back ten seconds. When if it's your bridge that's what it's 'remembering' to use as a guide for the next segment.
OK ThanX @neat nexus . I'm probably going to play around with substituting other ai singers for some of these tracks as well.
(just to add)
the 1st clip has a max of 40sec... this seems to be a hard limit to how much the AI can currently process in 1 chirp.
Continuation works by sampling 10sec from the previous clip, hence Continuations are limited to 30sec
(10 + 30 = 40sec)
If you're going to edit stuff together any way, it may be worth going to an earlier junction of the song and trying to continue from there as well, to give the later edit better continuity. As well as fiddling with the continuity of the bridge itself. It's an adventure in an unmapped country. So you're going to have to play around and find the work flow that works best for you.
That's what wetcircuit suggested.
Then how does the chorus remain similar from chorus to chorus?
And it's good avice.
How can I post my song from one channel into the showcase channel?
just download it and then upload it.
Hahahaha....if you're always getting similiar choruses then you're doing better than I ever did. I had to fight for consistency in my choruses.
Sometimes things go easy but you'll find that's a gift.
I was happy with the second chorus; there was a little variation.
The third...well I'll post it and you can listen lol
But actually I kind of like the way the third chorus turned out
yeah, lol. I was gonna say, they are just 'similar' but it is totally a craps shoot
🎲 🎲
I posted it in showcase; give it a listen
I'll give it a look-see after I finish the one I'm currently in...
very fun!
First song lyrics I ever completed, and Lord knows I've tried over and over and over and...
😅 I seem to start out with a song idea, hear the music and change the direction of the lyrics, lol
And I'll have a whole page of lyrics – witch the AI ignores... 😅 but I like where the music goes so I just keep going...
What do you mean it ignores the lyrics?
The AI will straight up ignore a whole chirp of lyrics and do its own thing instead...
I had it do that to me on a previous song
Only had four lines entered.
Are you on a custom plan? I just bought 100 credits, but will probably get a larger plan. Just wondering how much they may be.
yeah, I subscribed.... and then I bought a bunch of additional tokens..., lol
yeah, these 'top up' are in a different category of 'extra tokens' that don't expire...
I'd like to do something like this with my songs:
https://www.youtube.com/watch?v=AEUu3heIJNY
This video was created using Midjourney Ai and Runway. And being my 4th video in the series and it being the month of October, I felt doing something Halloween base would be pretty cool.
With that said I hope everyone enjoys the video and if you have any feedback please feel free to drop a comment below. Always looking to improve my work.
#ai...
@broken lotusHow many generations on average do you use to make a song, and what's your average song length?
If I have a specific lyrics idea in mind (like a diss-track/joke about some topic/ typical 'genre' song) it ends up really short, like my favorites are between 1min40sec to longest ~3min...
by 3min the vocal-quality has probably drifted so far away from the original, it's not really the same song.... And I realized there is no 'official' length for songs in 2023 (this isn't an LP or radio... it's the internet)
For instrumentals, I typically go longer, 4-6min, as long as the core sound is still there...
Stitched together in Garageband, my longest is over 8min (regen'd from an early chirp and made a whole new branch for the 2nd half of the song)
@broken lotus aha, thanX
Do you know if there are any full tutorials on making a good song with Chirp?
🤔 not really.... there's plenty of 'getting started' but... other than discussions here, based on trial and error....
probably the best teacher is trolling through other people's Suno videos in the https://discord.com/channels/1069381916492562582/1083129341870878780
seeing what they got, and didn't, from their prompts.... stealing their ideas (lol) and trying my own variation....
It's not bad. Seemed a little faster paced than what the lyrics were going for. In those cases I try to spread out my word count and do multiple short verses to make it easier for the AI to get the rhythm I'm going for instead of "hurrying" to get through the whole lyrics in the forty seconds or less it has to work with.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
My songs cut off at 40s. Is that the limit for free chirps, and if so, how do I upgrade? Thanks
Upgrading gives you more credits/commercial-rights but song lengths are the same. You build a longer song out of smaller clips, hit the continue emoji, add a new section (30s more) and then the heart to save the full clip: #⏩┃continuation message
@untold briarThanks
How can I put a guitar solo in my metal song? Can't find any prompt that works
Here are 2 examples when it worked for me....
(worth noting the 2nd clip in these chirps did not do the guitar, and the lyrics aren't exactly working here either)
@broken lotus thanks, so, you just put [lead guitar riff] in the lyrics section? After a [chorus], I tried [guitar solo] [instrumental break] etc. but it just cuts out.
agree there is some sort of (weirdness or just chance) with whether the prompt leads to a full instrumental clip or a cut short 3sec "I'm done!"
It's why I mentioned the other clips in these 2 chirps didn't do the thing I wanted... 🙁
I've also seen (someone else's) really simple prompt that just says the name of the instrument, nothing else.... (that didn't work for me, but worked for him...)
@broken lotus thanks, I'll keep trying 🙂
Jonathan Fry has a technique where he just spams the genre text field with the intended thing...
I tried this with specific instruments, and it usually worked (a la: "Piano, Piano, Piano, Piano, Piano, Piano, ..."
And let it come in where it decides...
What are the list of all the terms you can use in the lyrics, like stuff you put in between []
For example, what if I wanna do a build-up in one prompt and a beat drop in the continuation?
Wow! HUGE news with the expanded credits and subscriptions!
I was just trying to figure out how to pay for my new obsession. Now I can buy a yearly plan!
1000 chirps per month!
The cool is just beginning friend 🙂 Sooo many awesome things to look forward to. Very early days here
To do an instrumental break, I assume it's: [instrumental] all by itself in the lyrics?
Why did I get only two-second (silent) generations for one of my chirps? For this prompt, it was the second Chorus. For the second Pre-Chorus, I went back and generated it off Verse 1, which I had already continued. Does that make a difference?
Two things come to mind for this:
(A) if you want the entire next generation to switch to an instrumental you should be able to just add the genre you are using + instrumental break to the “pick a genre” section but adding an additional [instrumental break] to the lyrics section won’t hurt as shown in the attached image
(B) if you are trying to do a beat break into more vocals I would suggest keeping the genre section exactly the same and in the lyric section add [instrumental break], then skip a line and then add your next lyrics
Hope this helps 🙂
Could you provide a screenshot of what only produced 2 seconds, maybe I can help with that too
@daring lava Not quite sure what type of screen shot/info to include, but this is in "chirp-beta-7" about four or five generations up.
Thank you very much for your help! I'm heading to grab some dinner. Be back in 45 minutes.
Try [chorus]on its own line, instead of Chorus: -- could be the empty space after the : is confusing it, looks a bit like a script/musical/opera perhaps CharacterName: Song Line but then it's empty space
I can't get my second clip to be like the first one
The first four lines were so good
I wanted to do it like that on the chorus but I don't know how to make it do that
Can Chirp accept genres including decades? For example, [80's Pop] [60's Rock] [90's R&B]? What about more obscure: [70's Merengue] [2000's Folk]?
In my experience it does somewhat work. I think when I tried it was just "80s pop" no ' but not sure. I often kind of just spam keywords '80s pop, pop, 80s, 80s, 80s pop, eighties pop, 1980s if the simple version doesn't work, and often THEN you can get it.
Not sure about 2000s Folk, I think "80s SomeThing" was what I actually tried, and my guess is 80s is more strongly promptable than 2000s.
I just regenerated the clip using your suggestions, and it still generated two empty two-second clips.
I listened to previous clip and pretty interesting case I think. The way the singer sings that last line of lyrics has the feel of the very end of a song, doesn't it? Especially that "To light the way" really feels like an ending to me
So I think the AI is starting off the next clip sort of in frame of mind of like "I just sung the last line of lyrics of a song" and so it decides to just stop singing. But stopping is not a fixed thing, it's a percent change in these models.
So you might have to add extra lyrics in front of Chorus, and kind of hope to get a roll of the dice where it keeps going.
Glad to see it worked out #🐣┃chirp-beta-7 message I guess this ai stuff sure can work like us, sometimes you gotta tell them two or three times 😂
Btw cool song 👍
I generated it off a different clip
Yea, but you made it work 👍 Very early days here, but again, good job working it out to continue your flow
Yeah the clip is at a perfect stop, as if the last second of audio is the sound of the singer putting her microphone down on the table. So Chirp is like, okay, we're done.
A future Chirp could probably rescue this (for example a feature to force a minimum length on the output, or maybe even just turning up the 'temperature' could do it) but for now it's probably beyond reach without burning all your credits for a 1% roll.
If I generate the second chorus off the first chorus, will it be more likely to follow the first chorus' pattern?
If it's within the last 10 seconds of audio, yeah.
Also I rolled it and got a continue #🐣┃chirp-beta-7 message
So not 1%, more like 5% of not stopping on that clip. It did add a huge pause bit it DID finally sing.
I just saw that. What was with "country"?
That was just me throwing out ideas, I'm not sure if it is related to it not stopping though. Only 1 of the 2 clips didn't stop so could just be a coincidence. I am curious but not so curious to try 10 more times.
Is there any particular difference in generation style of beta - 1/2/3......
I mean, has anyone noticed any difference
Exactly the same, just to handle volume of messages using separate rooms. Maybe in future some fun rooms that are different though.
Thanks Jonathan
Has anyone tried [Interlude]
This song's going to be a nightmare to piece together. I've got it strewn out for miles on the channel, changing where I start my generation and which section I'm generating. Assuming I can find all the pieces, then I need to generate subsequent clips for those that have been cut off.
I made my own server to generate subsequent songs, but I liked this one too much to leave it stranded.
I guess I can edit these in Davinci Resolve...I need to learn that now for my YouTube channel anyway...
Don't update Davinci to 8.6, if you haven't. The audio is super buggy, almost unusable. Though there is a patch today... maybe it's fixed. Maybe.
I haven't d/l it yet
Should have done that today, but had too much fun Chirping 🙂
OK, now I got my whole song generated...just a matter of finding all the clips
You can also find clips on https://app.suno.ai/ though it's fairly barebones at the moment, not necessarily easier than Discord searches
aha I forgot about that, thank you. Saw that earlier today.
haha that's harder than through discord with so many clips
@untold briarHow can I complete this clip? 2 1/2 words got cut off.
Can you link the exact clip?
It may not be easy to complete if it cut off mid-word. But if it cut off at pause, then maybe you can continue it with the remaining words. But sometimes it doesn't join up correctly.
There a lot of ideas for future Chirp features, so at some point hopefully it is easier to do a tiny fix, like 'inpainting' an image, just one section, you could maybe redo just the last line of lyrics. But that's for the future and for Suno to make actually work. So it could be that you have to just redo that entire segment (recontinue it from prev) and get one that does finish.
It cut off in the middle of a word
I'll probably generate one and splice the last three words
chirp
You can also work in your DMs if that's easier!
Hey 👋 Go to #🐣┃how-to-chirp to learn how to use the /chirp command, then make your first song in #🐣┃chirp-beta-1
So I got frustrated that Chirp completely disregarded a line in the song but as I listen to it the more I wonder if it works better without that line?
Still annoying but sometimes there are happy little accidents?
Bob Ross mentality 💪
That feel when you realize it is the steel guitars that are making the hissing sound...
Ugh. I thought they were getting me the sounds I liked, instead turns out they may be the source of the noise I'm trying to be free of...
Oh wow. Restarting was absolutely the best thing I could have done here. So much better this time!
I was working in DMs for a while, but for some unsmart reason decided to do a couple publicly. These turned out to be my best efforts yet, but it was difficult tracking down all of the correct versions.
Pretty common question here - Can I run chirp on a 4gb 1650?
Chirp is Discord-only
musicgen is nice but can it do lyrics ? chirp is pretty awesome for what it does. will they add options for longer songs and etc in the future i hope?
You can make longer songs now 🙂 check out #⏩┃continuation to learn how
how do you do instrumental only? Maybe I overlooked somewhere..
Hey @chrome mango welcome to Suno! In the describe the music section just add instrumental or beat or something similar alongside the genre of your choice. Leave the lyrics section blank. If you bump into any issues after trying, let us know here and we will help better-refine your prompt to help capture what you are looking for 🙂
hmm what dataset is chirp based on?
The music sounds amazing clean
but the text to speech is the flabbergasting part
have a play with https://discord.com/channels/1069381916492562582/1108425425199308882
it's pretty crazy...
Has anyone ever been successful at getting anything Enigma like out of Chirp? Or even Gregorian adjacent?
without actual lyrics...? no.
Had a few un-prompted satanic sounding chants...
Nah, I meant with lyrics and gregorian type chants and so on...
I keep getting close then it fights me. 😦
interested in hearing a bit of how close you got...
It starts out really slow. Think Mea Culpa.
And it just dies there.
Subsequent attempts to goose it along are rendering silence.
So the only way forward will likely be to give up on the excellent chorus and try again. 😦
Or see what I can do with editing?
Majority of verses (and English translation) found here: https://www.prayinglatin.com/divine-praises-adoration/#TeDeum
Latin prayers of Adoration and Thanksgiving the heighest forms of prayer. Prayers in Latin with English translations and explanations of the Latin prayers.
🤔 pretty sure the original Enigma tracks were all editing....
Oh yeah.
Those would have to be.
But that was over thirty years ago, surely with Chirp we can do better today?
Also...
🧓
Just realized that that was thirty years ago...
yeah Chirp hasn't given me a single Whitney Houston vocal yet, and I even looked up the word 'melismatic'. What gives?
Looks like it's Steely Dan time...You go back, Jack and do it again....
The intro is a bit long any way.
And I want to see what I can do with groans and 'Ohhhhhhhhhhh'
'unng ooff bahh urff'....
May I quote you on that? 😉
I spent like 40min trying to get local LLM and then chatgpt to come up with scat vocalizations... bebop vocals, nonsense syllables.... (unsuccessfully)
Famous musical segment from the 1962 episode A Date with Jet Screamer featuring Janet Waldo as Judy and Howard Morris as Jet.
That one? It's the second result using this for style: Gregorian Chant, male vocals, chant
Argh. Now it won't give me music. Just vocals.
thankyouthankyouthankyouthankyou
thankyouthankyouthankyouthankyou
I see it worked out better for you than it did for me.
you are too picky, lol
I let it go on a journey.... from arabia to bollywood with a detour through hollywood-celtic... lol 🌍
I'm starting to wonder if that's part of my issue, also if I need to stop trying to write stuff ahead of time.
I seem to get really good results when I just 'doodle' and let things go where they will. When I try to force structure on it, that's when I tend to have issues a lot, I've noticed. Because in the writing I've got an image in my head.
💯 👆
None of my songs end up matching their initial lyrics lol 😭
and I'll regenerate if both clips show Chirp wasn't done with that section... try to write new lyrics that fit.
Also that above song was a lesson in breaking up words to get the AI to speed up or slow down...
and I let it ad lib a lot, lol.... But it's Ancient Sumerian translated phonetically. No Latin scholars are going to get mad at me....
My Italian friend on a different Discord really didn't like how the AI pronounced its latin. LOL so yeah, it's a thing.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
This is wild!
oh wow if it wasnt for your reply i may have missed it. ideas like that I sure hope vocaloid songs come back in style. i used to listen to Crystal Castles - Kept all the time
as in jibberish samples pronouncing something that resembles words but not
Does anyone know if its possible to only have acapella voice from chirp ?
yes. use 'acapella' in the genre description.
Thx for your reply. I tried but its doesnt work. Maybe because i try to have "scream" or "metalcore" acapella ?
I have not got a scream to work in Chirp.... 🚫 😱
Prompting a genre that does NOT normally do acapella, and instead has a very distinct (overused) voice that it considers appropriate for that genre, is sending competing prompts to the AI.
Maybe instead of "metalcore acapella" (notice which word is first?)... try describing the quality of the voice (angry shouting etc) ...
Or try "metalcore as acapella"..., "metalcore but acapella"... I do this with instruments: "Funk but tribal instruments" and it works...
I hit a similar issue with trying to have a female singer in hard rock.... I couldn't get around the AI's bias...
(it just gave me a falsetto male voice)
Very interesting ! I was not sure about the prompt, according i was completly impressed by the song quality generating a metalcore one. So if i try "angry shouting 165bpm" with my lyrics, you think it works ? 'Cause i never used "metalcore as acapella" which seems too interesting.
😬 🤷♀️
I mean..., I haven't tried these specific things.... so... It's just a suggestion...?
Oh alright i see. Well i went to be sure because as i have a free account i dont want spend all my credit (ahah). I have a serious project to do if it works :p, but thx for help, im gonna try your suggestion
If it works, let us know! 😃
(or update if you discover a good trick)
I sent an email to support yesterday for this information. I ll post there once they replied
::sigh::
It's amazing how nice stuff started flowing again, once I gave up writing BS and started being real....
Anyone have suggestions on how to calm the AI down again? I've tried my usual [tempo shift] and it does nothing.
Slow, Slow, Slow Slow...
?
If this next one doesn't work, I'll try that. Thanks.
Sadly that didn't do much to turn down the heat.
I thought I had some luck with [backbeat] [instrumental] but not sure how much it helped really.
I don't think 'backbeat' is the right word...
found this discussion:
https://music.stackexchange.com/questions/78320/what-terms-are-there-for-changes-in-tempo
I guess I need to get Ritardando? 🤌
Here's my WiP and you can see where I'm having issues...
ok, well one advantage is the singer hasn't started yet..., so I would spend a 'chirp' slowing down during the instrumental...
[Percussion breakdown]
[Break]
[Italian R-word]
...
even bringing the song to a near still before bringing it back up...? (song will probably completely change in that transition)
Ritardando did exactly what I wanted it to do. Now let's see if I keep the voice for the next verse.
See this is what is nice about Chirp, even if I don't know the musical words, it does.
So once I find the right words, it unlocks things.
Italians are to music what the French are to underwear.... When all the words are in 1 language, they were the people who invented it, and actually used it....
😓 should have studied Italian in hs...
Man it just hates this song. Every generation is now barelt 12 seconds...
Add more lyrics?
That doesn't help when it's wildly interpreting the voice when I do that. Will try nonsense sounds to see if it helps.
When I'm trying for length... tbh... I just start previewing the last 10sec of the clip... since that's what matters for the next continuation...
I've been getting four seconds if I'm lucky. Just had a brainstorm though, waiting for generation on it.
(as in, 'is the voice still coherent in the last 10sec, without buzzing or ringing or sampler noise')
Well the brainstorm didn't pan out. I tried [scat] to see if it would do anything. No luck.
Not sure if it might actually work with some different wording or not. Was an idea.
Maybe [vocal scatting] might work on some other attempt?
oh yeah... no scat.
[Ad Lib] sometimes works, but nothing beats a big text file of Ancient Sumerian for slapping down those groovy rhymes mid-line....
I know about [adlib] or does it need to be spelled [ad lib] to work?
I spent half a day trying to arrive at 'bebop phrases'.... tried everything I could think of....
ChatGPT came up with nursery rhyme scat... LOL
Sure, here are 40 nonsense alliterative phrases for a bebop singer:
Bop-a-doo-wop
Fizz-fuzz-fizz
Goo-goo-ga-ga
Zip-zap-zoo
Jibber-jabber
Quirky-quack
Razzle-dazzle
Tickle-tackle
Yip-yap-yo
Boogie-woogie
Doodle-dandy
Hocus-pocus
Jive-jangle
Zig-zag-zig
Ding-dong-dell
Pitter-patter
Wobble-dobble
Rumble-tumble
Whirly-twirly
Zany-zigzag
Flim-flam-flum
Wobble-wabble
Fiddle-faddle
Twizzle-twazzle
Doozy-doo-doo
Ziggy-zag-zog
Tootle-tootle
Riffle-raffle
Quibble-quabble
Zipper-zapper
Whammy-wammy
Dribble-drabble
Snip-snap-snazz
Zippity-zappity
Quirky-qwerty
Wiggle-waggle
Doo-dah-dandy
Skibber-skrabble
Whirly-girly
Hubble-bubble
Damn. That's much better than my attempts...
I keep meaning to get back to this one some day too
This is how an AI says 'I'm bored with you right now.'
Notice that there was nothing entered to make it do this but it started a slow fadeout.
See?
No commands used to do a fade.
Chirp just basically gave me the raspberry.
lulz
Well back in the vault it goes until it can be released...
Now I need to make something else to prove that I'm out of that slump.
Really bothered me that nothing was happening these past two days.
Does anyone have any idea how many users Chirp has currently?
Whoops wrong section...
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
has anyone had problems getting the full song versions after liking a chirp, if the song is around 10 minutes or more? im out here vibing, lyrics flowing, song turnes in three different ideas, i follow them all till the end and cant get the song. it happened before, but i got them a day later, now its been a day and nothing. its ok if i have to wait it out, just not if i now have to stich together individual chirps trying to recreate.
more broadly, does me liking every step of the way guide the AI in any way, or should i leave the like for when the idea or a song is finnished?
sounds to me working as intended. you only get 10 chirps a day and thats counting test runs on stuff you dont thumbs up
yes, I have a few that I can't " ❤️ " at the end of a chain...
I can get a few Continues earlier, so there is some issue or cutoff
I don't think you are effecting it by " ❤️ " along the way.... But idk.
according to discord splash page....
Hey guys how do I chirp with just the bot like in a PM so no one else can see my lyrics?
just DM the bot like it's a person, then you can chirp in private message.
k ty
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Okay, I guess I could have figured that out someday
Certainly! Here's a revised version of your sentence:
How can I search for songs that have been created for me, including those created in the past?
SEARCH ...
MENTION: USER....
Hi friends, I've got a problem. The issue is that it's coming up with lyrics I haven't written. What's the solution?
And it's using up my credits.
Hello
Does anyone know how to increase the duration of the song?
see #⏩┃continuation
perfect, tks
write 'instrumental' in the type field? what are you trying to get?
a song with my own lyrics
hm, then it should be fine. does it continue after yours end or just sing different lyrics?
sing different lyrics before my own lyrics and sometimes dont sing my own lyrics at all
if I write intrumental it will even delete my lyrics too NO?
he sings different lyrics before my own lyrics and sometimes he doesn't sing my lyrics at all and makes up lyrics
can you show me an example? is this english or different language?
strange, can you send an example?
Is it possible to download a gen in an actual audio file?
Not yet. Developers say it is a goal for the future.
try ffmpeg -i filename.mp4 filename.mp3
I can convert to mp3 no problem, I was hoping for wav or flac
Just a tip for people who might not know: If your lyrics run long and get cut off midway due to duration limit, just repeat that verse in full via "continuation" and the ai should continue where it left off to finish singing the lyrics without repeating everything.
Is it possible to download the vocals and music separately?
You'll need demuxing software like Ultimate Vocal Remover https://github.com/Anjok07/ultimatevocalremovergui
Is there a chirp bot to DM alike in midjourney?
Sure. Just DM this guy:
Thanks! I've had decent luck with Moises on the Mac to extract vocals.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
hey
I need help
i wrote the whole lyrics but then it is only genearting 1st verse only
#⏩┃continuation
i use /chirp but it doesn't work
hey all, not exactly asking for trade secrets, here, but wondering what software is being used for chirp... I see there's a repo for bark but is there one for chirp?
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
When I type /Chirp and hit enter, I don't get a popup. Nothing happens in Chirp-beta-1
🤔 just type /
and a popup should happen with various commands... maybe it has to be all lowercase?
You may have to scroll down and accept the Terms of Service. If it's getting lost you can try in DM with @versed valley
I just bought a pro license for a year. Where do I go to see my added benefits and faster processing?
Type /info
Well that shows credits, you don't have to do anything differently for the other stuff, rather, the bot just won't stop you when you would have capped out
I have the pro plan and it says active in manage accounts but showing free here
Priority generation queue
10 concurrent generations
My guess is there's just a little bit of lag in Discord syncing up, if you just subbed
ok thx
But message mc from Suno if it's not showing the accurate numbers after some time
hey checking in on this, DM me and we can get it sorted out (there's a chance you may have subbed on the web on a different Discord account?)
No problem, we'll switch it over for you to your main email. Standby! But I'm gonna delete your messages above so it doesnt show your emails
Hello,
Is there way to generate a video song from the lyrics?
So like....I don't know what this was but it had nothing to do with what I put in there. Is that even English?
is that a Continuation...?
Chirp: How about we make a Love Song instead?
🎵 mlenginubbofleazur–LOVE–ingrubnyfurkom...
Part of several attempts at a song. Not having much luck despite really loving some of what I was getting...
It starts off good then either noise creeps in or other issues.
Obviously these all start the same then diverge.
this chirp is cursed...
And I see the 2 first songs introduce jibberish singing at that same point...
Yeah, that was in part why I gave up on them after a bit. I kind of liked some of the gibberish though. It was alright when it kind of worked. I figure that since no one sees these videos but us, that it would just seem like scatting or adlibing.
I'm willing to go with the flow because sometimes it takes us to good places. This today though? Yeah this has been rough...
I still think there's some good stuff in there here and there though.
Wait that one? That specific section or just the whole song of that chirp? I kind of liked that section before it derailed.
that's the chirp that introduces noise and starts degrading the overall sound (in my flippant, uninvited opinion)
That's the most recent attempt.
They're named in order of attempts.
I have a whole folder I created for this series of failures called bad luck, where I went back to earlier parts to try to resume without the noise or other issues.
You're saying that segment there is where the noise is introduced in general?
So I went back up and took the other fork and will try from there. If I still can't do it I'll have to pitch this entire effort and start over.
So this is the other side of things. Some interesting stuff here. I stopped because I think I heard noise starting again.
With better phones on it's not as bad as I feared but it's there.
is NICE
(but singer flipped to female)
I noticed that but it was such a nice vocal I couldn't leave it alone.
agree
I suppose the solution is to cut the segment where the issue creeps in and do a completely new bridge instrumental.
here's an example where at 1:10 (3rd chirp) I used
👉 [Melodic Instrumental] 👈
and both versions were so good I extended each...
On BOTH songs, when I'd had enough and switched to
[Bass and Percussion Interlude]
which wrapped up each song satisfyingly....
Oh man...the one on the right....
swoons
I'm going to resume from before the issue and steal that melodic instrumental from you and see what it does.
Is there a way to have more control over the music or voice? I feel like I waste a lot of credits because I don't like what it gives me..
actually yes. I managed to affect that by describing the voice in the genre
Well I think I'm giving up on this one. It has momemts. Some really nice moments. But I can't get it to turn back to the chorus. Either it renders it in hiss whisper or it ignores it, or just cuts itself at three seconds. I've spent enough tokens on this one I think. I'll save the lyrics for revisiting. I've had at least one or two good songs from doing that. But I think it's time to try something else.
I'll definitely remember the [Melodic Instrumental] for future use though.
yeah the melodic instrumental sounds great!
I request Country, Garth Brooks and it gives me two female voices. I feel I'm wasting a lot of tokens trying to get the chirp to give me what I'm looking for. How do you get around that?
Chirp is deliberately NOT trained to use musician names or Band Names because Suno don't want to be a 'covers' app and get in legal trouble.
You need to describe the voice, something as simple as:
Male Vocals
in the Genre.
If it's stubborn (like you're fighting a genre bias), repeat the prompt a fe times:
Male vocals, Male vocals, Male vocals, Male vocals, Male vocals, Male vocals, etc
as a follow up to the previous question, can we describe the style as particular influence, like Start Wars as opposed John Williams?
I asked for male vocals country, male vocals country and it gives me this...
😆
we fixed the gender descriptions for the next model. it's kind of a crap shoot right now unfortunately
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Might be an effect from the lack of comma after vocals. Maybe try male vocals, male, country, and don't be afraid to crank it up to something like male, male vocals, country, male, male, male
hmm it flags this for some reason:
style: german rap
lyrics: Frag mich weiterhin wen du hier wegboxt!!! Press dich mit paar' lines ausm' Tank-Top!!!
The AI moderation isn't that great. Hopefully better in future. It's actually interpreting this as self harm instructions, somehow... I was curious. One way to make it work as add additional lyrics, seems to change the context, even with no changes to your original line:
Hey, ich bin neugierig! Frag mich weiterhin, wen du hier einlädst. Zeig mal, was du mit ein paar Zeilen hast!
Frag mich weiterhin wen du hier wegboxt!!! Press dich mit paar' lines ausm' Tank-Top!!!
Here's a funny fix, add two emojis:
🎵 Frag mich weiterhin wen du hier wegboxt!!! Press dich mit paar' lines ausm' Tank-Top!!! 😊
lol nice
It's GPT/OpenAI they have a tool which does it. Not the most amazing accuracy especially in non English though 😄
i used Perspective API for some discord bot, but it's also not perfect, specially against the weirdos typing something p**o it fails hard
I don't know if the emoji actually works for Chirp, I was testing on OpenAi directly, possibly those symbols are filtered out
maybe if instructing the AI somehow that it's for artistic purpose where moderate profanity/disses would be allowed or so? ^^
Yeah I think that's a possible good choice for future improvement. If you ask ChatGPT 4 to analyze lyrics it's quite good. But much more expensive of course. Maybe 3.5 alone could be better, not sure.
maybe it just hates german language tho 😄
Funnily enough I asked ChatGPT "Is there any way to interpret this German message like X and Y (the stuff it was flagged for) and it was like, well, it's a bit violent imagery, but basically no, seems like other AI screwed up. lol
haha
seems chirp struggles with text containing emojis, or the text is just too short, or german is the problem ^^
Same as Bark. Very short inputs often have high chance of problems. I think it's simply because most of what the AI sees during training is longer audio clips, it's kind of an edge case.
But for moderation, also true I suppose, because less content = less context. Adding one extra line of German did make the basic mod AI no longer see original line as self harm. Maybe it's similar.
^^
thought the style maybe was the problem so i tried to see if this marks it maybe if it's also weird context, but works with the random sentence in lyrics now anyway
many thanks ❤️
I clicked your profile and went to your github to follow and I already follow you... did we talk earlier? lol. Or maybe i was just following anyone who forked bark repo
uhm, i think we had some convos some months ago already and we talked about bark notebooks iirc
i had this weird notebook which trims out silence
thx for remembering to follow people back on github tho
Lately following people on github more fun than Twitter, makes the Github feed fun to check every day
never did that but will do in future, always thinking about searching for cool projects but then i forget about the prob inspiring feed of people one follows
....
Man I really need to let this one go.
I went a bit more carnally with this attempt and did different verses as a result.
Still doesn't quite work. It dies about here because when the music ends it doesn't come back, for nothing.
I thought I maybe had a bit of a love song on my hands for a minute.
A bit of a carnal one, but still.
😳 That ending is sooooo dramatic tho.
I like it! 👍
It's like a segue song on an album. Next track starts BIG, lol
Should I keep it then, and see what I can do with it in terms of placement?
Not sure if it fits with my other stuff.
hashtag: hidden track, hastag: bonus track
Any thoughts on which one is the best to use?
Melodic as a term makes such a difference ina lot of things.
How many verses and choruses can I add to one prompt? The first clip came out amazing, but when I tried to add the second verse and chorus with another clip continuation it sounded weird.
Also, for those of you putting male vocals, country, etc, do you put that in the style box? Curious how powerful that style box is.
How do you share clips like that in here? Just download them and paste them in a message?
Yeah, that's exactly what I do to share.
I like 2, but it's too short to be of any use. And even if you combine clips they a lot of time end too abruptly.
Any timeframe for that release? Also, even when combining clips, they end abruptly. Is there some kind of prompt I can do where it will actually end the song proper especially if combining two or more clips?
no release date yet but training it now. you can try things like [Outro] to end a song
Yeah you're right. I found that one out the hard way. 😦
Okay, thanks. What I wish the bot would do is especially when using the random option for style, if once generated, told you the style generated. I'm running into a situation where I have an amazing clip I want to continue from, but when I put the [Verse] tags with the new verse and [Chorus], it goes off the rails.
if the Continuations are doing something similar to each other... change your lyrics to fit that...
Sometimes easier to hear where the AI is trying to go, than to fight it.
But I'll rise from the ashes, I'll find my way,
Though my heart is heavy, I'll see the light of day,
Broken, but not defeated, I'll learn to be okay,
In the ruins of our love, a new me will be born today.
See, this here was the result of combined clips, but once you combine clips into one you can't continue from it. But I'd love to add an Outro to this.
Why can't you continue from it?
Whichever one you hearted to make this one? Continue from that option and add an outro.
It doesn't give me the continue options from the combined clip. It says I liked a clip, gives me the full clip with no way to continue from that. Unless you mean continue from the fragments.
Yeah. Just go back up from the last part before you combined, select the fragment or segment that you used the heart on previously, only this time continue it with the outro you wanted.
speeling matters.
Also the place for that is in one of the chirp-beta channels.
how can I search by keyword only the fully generated songs?
Hmmm...first part, solid intro. Nice.
Second part, oh I like the instrumentat---WTF?
Chirp: 🤸 🤛 🤾♂️
Is there a mark stressed syllables on words? Chirp usually does a good job, but not always
Use all-CAPS where you WANT the accent
I like how if chatgpt fails it'll just put the error as the song 💀
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Is there currently an outage? for like 10 minutes none of my generations have finished
nvm it just finished
I'm having the same problem, except I have generations that are 53 minutes old that haven't finished. Looks like if I start a new chirp, it will finish it, but if I continue anything, it never finished the generation.
This one is still showing "Generating..." from 8:10pm (57 minutes ago). Thats the oldest one so far that hasn't finished. 4 other continuations are still sitting in queue with this one.
yeah what I usually do is, if one chirp takes more than 5-10 minutes, I'll make another one, and usuallly that one finishes
That's what I usually do, but they aren't finishing for me it doesn't look like. The 5 jobs, including the one from 8:10pm are still showing "Generating...". Hopefully, Sunos will get over its electronic snit with me tonight. I'd hate to hit a sour note and find out it's composed a grudge against me by tomorrow.
So obviously AI is AI and that means there are bugs and various issues that need to be worked out. But something I noticed that is interesting is how often AI will scan a song made by AI and conclude there is no music there. Sonos Teller will do it sometimes, and now some bot on Reddit's AI Video sub scanned the video and concluded the same thing.
Can you guys offer some insights on how this keeps happening or why it keeps happening?
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
it's called SNAKE OIL
the author of this 'bot' made the assumption that your song will be something already in it's database. When it can't find it there, it gives a generic error and says there is "no song". That's a red flag the bot is a scam.
Many similar bogus claims were made for AI Art 'detectors' which were not detectors at all, but reverse image look up databases. They would find any similar image in their database and claim your AI Image had been trained on this and this specific image stolen from online.... These 'detectors' were scams because you could upload ANY image (not AI) and it would still declare that you had stolen photos from the internet because they were similar to this and this in its database.
As the hype about AI moves into music, you well see MANY of these bogus 'detectors' that are just reverse search engines claiming to be magic snake oil. A false positive (in this case a false negative) shows they aren't doing anything useful.
I see.
That definitely makes sense for the Reddit one. The one in Sonos Teller on the other hand...sometimes it works very well. Other times it just fails.
That one is actually doing more than looking stuff up I think.
Even if it gets stuff wrong sometimes.
How can I share this discord channel on Linkedin? I'm writing an article and want to link it
Suno.AI is the actual web address. There is a permanent Discord invite there...
Thanks
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
things like 'genre' are not real. It's just marketing.
If the bot is suppose to label your 'genre' it just means your song isn't as generic as (whatever criteria for genre it has)
I mean yeah that is what I used it for in the past. But there have been instances where it just straight up insists that this is not a song.
LOL
If it's marketing then why is it useful for people trying to find more of the same?
If it's just marketing why does it help us in Chirp creating songs?
... that's what marketing is?
Kinda why vegetables all somehow end up in the same grocery aisle....
(but what do carrots and lettuce have in common genetically?)
I guess? But if I'm looking for a bunch of green onions I'm going to get frustrated if someone hands me a couple of leeks.
They might taste the same. But it's a different preparation to them each.
exactly, so what do they really have in common?
And I never tried eating raw leeks. Green onions in a bit of salt can be good.
After looking up bands to see what genre they are to tell Chirp to make something similar..., before the mid 80s there were no genres (well there were, but...) In the US record stores were segregated into White people music (Rock/Country) and Black people music (Soul/R&B). There was probably a lonely backshelf for 'the rest': classicaljazzmusicalsoundtracksfolkgospel
In the 80s, niche labels were able to ship through common distributors and an explosion of 'genre' happened, hitting peak in the 90s with hundreds of sub-sub-sub-genres.... until it became about music 'journalism' and niche marketing..., but meaningless from a label/distributor/store classification – they're all just alphabetized now.
But looking up those older bands, their genre descriptions are always a hodgepodge of 'crossover' because (prior to Disco – an actual genre) bands didn't really set out to be 1 kind of genre... They had musical influences but they didn't set out to be 'a genre'
I take your point.
But then again, since the whole reason I look up my own songs is so I can make sure I hashtag them correctly in YouTube for marketing purposes. So what-evah. 😉
Don't hate the player, hate the game.
Most of what people think as 'genre' is actually just a 'style'. Genre's need more than 10 bands participating..... and genres are a lot longer lasting than a couple years of trendy 'style' sounds.
It's acting really weird when it comes to flagging
I put in some lyrics and it works completely fine, but if I put in the SAME lyrics, with the style dubstep, suddenly it's flagged
There was also a time where I used the same lyrics, but capitalized some words for emphasis
After some experimentation, it turns out that the lyrics get flagged for every style, except "fast rap", and I don't know why
What's Sonos Teller?
It's an AI tool that takes YouTube links for songs (can be unlisted cannot be private or scheduled) and gives summaries of the type of music, genres, styles and instruments it detects in the song.
Here's an example of my latest song released on YouTube:
https://www.youtube.com/watch?v=PeMqQwGyGBY
This is Sono's summary on the song: https://sonoteller.ai/PeMqQwGyGBY
Notice though that it thinks the vocals are instrumentals.... 🙄
I notice that even after the the limit raising from 30 to 40 seconds, sometimes the song will cut abruptly, missing out on a part at 30 seconds, even though it can make it longer
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
it's 40sec on the first Chirp only.
Continuations are 30sec because they sample 10 sec from the previous clip (10+30=40sec)
when I have a line that spans across a Continuation... yeah, there's no sure way to make the lyrics seamless... Sometimes adding the rest of the line works, sometimes using the whole line works better and the AI picks it up where it wants... Sometimes the AI sings that line again but in gibberish (I don't mind a little gibberish if it's not too obvious/distracting).
Usually I don't Continue a chirp that has a problematic cut, knowing it will cost time/tokens to try to get one that works.... Easier to regenerate the one with the problem cut, maybe adjust it's lyrics shorter or just hope for a better 🎲 🎲
You nailed it. Have to try everything and also get a bit lucky:
My Chirp is missing the last bit of lyrics, can it be fixed?
Best Solution: Don't try to fix it with a continue. Just try to regenerate a new version of the cutoff clip that is complete.
I really want to try and save this Chirp with a continue:
- Try adding just the rest of the line, starting from where it cut off.
- Try adding the whole line, the part before and after the cut.
- Try adding the whole verse and then after your new lyrics. Works at least as often as 1 and 2. Chirp loves aligning things I guess.
- Even try re-using your entire previous lyric prompt - literally copy and paste all the text/versus you used 100% exact match. This sometimes works, like it can start at exactly the cut off word, like a miracle. Don't skip any tags you want a character-for-character, space-for-space identical prompt.
- Same as 4, but also add your new lyrics at the end of the identical prompt. So old-prompt-exactly + new song segment.
But also yeah, 9 out of 10 times faster and better to regenerate the cut off clip instead. But if you really fell in love with it, you can pull out all the stops.
Actually has anyone tried
6. ? Continue the song with no lyrics? Not very likely, but maybe if only missing a second at very end. stranger things than that do work. I'll try quick.
Anyone got a link to a cut-off song, maybe just missing one word?
🌈 my fantasy strategy has been to try to trigger clips that all end early – hopefully as seamless cuts.... like at 24-29 sec...
😅 this almost never works.
There is no way to convince the AI to end a few seconds early. When it happens it is not always the best clip, and when I try to do it on purpose the clip's usually TOO short....
my experience is if you catch a vocal across the cut, and don't supply any vocals in the Continuation, the Continuation will keep vocal Ad Libbing through the Continuation.... The vocal is burned into the music sample and won't shut-up.
Number 6 does work in Bark, but Bark has a stronger sense of meaning and real words than Chirp, despite being a terrible singer. But Bark can guess a word and a half pretty reliably.
(Chirp has a bigger brain but uses a lot of it on the complicated music stuff.)
BTW always interested in cool Chirp ad libs that actually have recognizable sentences. Or at least phrases using real words. Usually you get Simlish songs, also fun, but curious if they can make sense... anyone @ me if you happen to make any, if not a private work or whatnot.
Wonder if some [tags] or non spoken prompt content that makes it more likely to stop naturally around 24 or 29. Maybe not even necessarily at the end of the lyrics. Could be using 'padding' tags throughout song gives the AI more space to adjust timing (slow down, or skip over to speed up) and it nails the landing a bit better. 🤔
Any time I try a series of non-lyric prompts, Chirp puts them all together as 1 prompt, so they either cancel each other out, or possibly all happen at once and done in 3sec.
Does---
Ok, this is going to sound weird, but bear with me...
Does the AI get "bored" with prompts?
Like it gets the same submissions, tweaked but basically the same and it just quits trying or sees the additional submissions as errors or redundant so it just doesn't do anything?
Getting diminishing returns on a song I'm working on and it's getting frustrating.
When it's not just ignoring me entirely to repeat itself or making up stuff on its own it just generates noise. The spectroscope indicates sound, but I certainly can't hear anything.
Or it will just abruptly end the song segment in such a way that it cannot be restarted.
So like...is Suno Chirp bored with me or something?
waits for the developer to give the 'it's not you, it's me' speech
Completely separated songs no.
But something like that can happen in continued songs. Even if the prompt repeats a line that isn't in the past 10s of the clip, so you might assume it couldn't be considered redundant, the prompt can still have an effect on the audio overall that persists. Like in Chirp V1 how the music style itself was a determined by lyrics alone. So repeats could magnify artifacts or errors. (Bark hates repeated prompts for sure.)
Even in pure text models similar concept that quality tends to degrade by building up. Garbage is more likely after garbage, errors accumulate. Here at 28 minutes, ChatGPT has a super hard time recovering after it made a mistake, it just keeps making more mistakes. Only if you go back in time *before * it made the first mistake can it answer correctly: https://www.youtube.com/watch?v=jkrNMKz9pWU&t=1673s
I don't know if it's directly analogous technically, under the hood, but as a human person prompting the concept seems to apply.
If you are using brand new song gens, what's probably happening is that those prompts in particular tend to produce that kind of output. So if a particular lyric and style prompt keeps ending way too early, even though it should be plenty long, that could keep happening "over and over" just because for whatever mysterious AI reasons. Not because it's a repeat of an early attempt, but just for that specific prompt, Chirp wants to end a song real quick. (Do not question the will of our future robot overlords.)
The first posted one, about a minute in, the second attempt about two minutes in.
In both versions you can see the behavior though. It just skips lines it doesn't care for.
just a guess: maybe if you omit the style prompt on a continued segment that keeps skipping lines, that might reduce the chance of skips. Without a style prompt it could pay more attention to the lyrics. The reverse is true, if you super overload the style prompt, it for tends to forget about the lyrics more often.
Hmmm...
You may have noticed in the extreme case, Chirp will actually sing the style prompt instead of the lyrics, lol
I tried to avoid doing that, but occasionally it seems to help if I goose it along. Especially switching between the vocals and the musoic styles.
I may have encountered this a few times. 🤣
I think the style prompt used to have a longer max length and if you just packed a novel in there it would almost always do that, haha.
Well I completed a version of the song at least. Probably will come back to it and see if I can goose it closer to my thinking.
Iterative song making isn't really a thing in Chirp but sometimes you can process something close to it.
And a radically different style can fix a song that was just alright into something much better
I have found if you have quality degradation from too many continues, sometimes you can competely 100% 180 the style prompt, and recover, but change music genre. Yeah wow you typed same time.
It's not really a great solution since generally people want a song in one style, but works if you can do that.
Yeah.
I meant also that sometimes just doing it completely differently can have benefits.
This for example wasn't bad but I think the remake was much better overall...
Jaxon Rhythm made this simple little song about a man who wants to forget about the past and just enjoy a day with his girl. Something I think a lot of us ca relate to, honestly. No need to argue or focus on past hurts. Just love.
#demotape #experimental #rock
It's the same lyrics, different vocal choices and radically different music style.
Lyrics comparison:
Is there a style prompt or something to make Chirp make a smooth ending to a song? I did an Outro but it just ends abruptly. Unless the Outro metatag didn't work and I have to just keep retrying.
Is there anyway to have chirp give you a complete clip if you like the initial style? If I click on the heart emoji, all it does is just generate another 40 second clip that cuts off at the end.
You click the continue button on the first clip, that make sa new 30s clip (with new lyrics.) Then click the heart on the that clip (or the last clip) and bot will combine all parts to make one long song
That's awesome, but I really want a way for it to smooth end. There's no way to have it fade out, no way to have a little musical end after the vocals or something just to make the ending less awkward. It just cuts off like a switch.
You can try using a short lyrics prompt, or empty prompt, and then ending with something like [fade out] or [song ends], though you have to get a bit lucky to work
The problem I am encountering is that if I click continue, it only gives me four more seconds, and asks for more lyrics when the first clip only used a third of the original lyrics. Just so I’m clear, I have to keep continuing four seconds at a time until I get to the end of the song and then click the heart? I’m going to try that now, but it would be great if larger chunks could be generated on each continuation. At this rate, it’s going to take forever.
OH no, if you're only get 4 seconds in a continue, you might be out of luck. That could mean the particular clip is kind of bugged and keeps ending too quick. A continue can be 30 seconds long. If you keep getting 4 seconds then either 1.) The clip is bugged. 2.) The prompt used in the continue is somehow making it end too soon (no lyrics for example.)
Add the unsung lyrics to the Continue chirp....
(split up the lyrics across your Continue chirps)
its jover
I’m experimenting with this quite a bit now. With each continuation, the style completely changes. Keys are different., it may be a female voice instead of a male voice….. even using the same exact prompt doesn’t create continuity. Overall, I think this will be a great tool when it can maintain context.
I'm trying to end the song on a continue. Can I do [end song] in the genre box?
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
When pasting lyrics, wil [Verse 1] work as a metatag or will the number be ignored?
it shouldn't be sung allowed....
I've copy/pasted lyrics where the prompts were just separated by a comma, and Chirp figured it out
(other times I have heard chirp sing the prompt. especially a long prompt it didn't understand)
What is the best way to do Metatags? For example, if I want a 1950s country ballad with a female vocalist and steel guitar, the following seemed to work, though not consistently.
[50s], [Country], [Ballad], [Female Vocals], [Steel Guitar]
Not sure if it's best to uppercase the tags or what.
Too bad I can't do this with Chirp...from ChatGPT lyrics:
[Bridge]
[Instrumental breakdown with heavy guitar riffs, pounding drums, and an atmospheric undertone]
Fails every time. Guess I'll have to skip that.
Is this song raw Suno? Or did you edit it and add stuff? Pretty impressive.
Which one are you referring to?
@neat nexus Depression.
Depression needed edits. The chorus was such that it couldn't be replicated. So I basically had to splice earlier pieces together with a continuation from other earlier peices to maintain consistency.
Sometimes a song will just work from beginning to end with raw Suno Chirp, other times the only way forward is back.
No, that I sourced from a video. The source is mentioned in the details on the YouTube link.
While I was told it was public domain, I still felt that crediting was a good thing to do.
The Suno docs say there's a Made with Suno channel. Where is that?
It was split into chirp-showcase (made with chirp) and bark-showcase (made with bark)
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
/chirp
You need to be in the chirp-beta-2 etc channels
tune ur noodle 😄
A Furious Chirper suggested to me this might make Chirp less likely to cut of your song in middle of a word. (Not trying to end the song itself like a [song ends] or [fade out] but just end the segment cleanly at 28s, end of last line of lyrics, instead of 30s in the middle of a word on the last line.)
This is the last line of lyrics...
(maybe a blank line here?)
.
.
.
|
[optional descriptive of some sort here too] or if you want to end song try [outro] [song ends] [fade out]
That's three periods on lines by themselves, followed by a pipe character | on last line. And optionally a descriptive sort of tag at end.
Seems plausible to me, a flexible buffer at the end, so AI has margin for error, the last line of lyrics is not right at edge of prompt.
I haven't tested this yet, if anyone does let me know what you find out!
I will say the pipe | character is actually weirdly powerful in Bark. So I would not be shocked if pipes are somehow cool in Chirp, or have some effect stronger than other random punctuation.
In Bark it means 'a new person is speaking' for example. Hi, I'm Suno | Nice to meet you, Suno! would give you two voices in Bark maybe half the time. And as far I know this is not by design, just one of things Bark picked up.
I always add blank line after my lyrics.
I sometimes use an ellipsis (...) to add a little empty lyric space
Never used a vertical bar/pipe...
will have to try it
Blank lines makes sense, yeah. I'm not positive but I think multiple blanks are collapsed into just one, so maybe one line maxes it out. So could be adding periods is kind of like cheating your way to 4 blank lines.
yeah, that's how I've used the ellipsis – as an unsung vocal space
(also works at the end of lyric lines to make a 'rest' so the vocal doesn't try to eminem through to the next line.
I guess this is a technical Chirp thing. Was getting mad discord lag when bouncing between back and forth continuing old segments (way back in chat.) Just tried this "reduce motion" and it seems to help. Not fixed but better.
Is it possible to pass a pronunciation phonetically? Like using the dictionary thing? Because it does not pronounce "gliding" correctly.
I actually have no idea. I'll test it for you so you don't need to burn credits if you can give me a few prompts. Would be useful if it did. (Low odds, I think though.)
The word is gliding. It wants to pronounce it Gl-id-ing rather than gl-iii-ding.
The words in context:
We come together
we meet in the middle
but we don't mix
oil and water
oil and water
just floating on top
No depth to be found
Gliding around
oil and water
oil and water
but we just don't mix
I did in fact try doing it as Gl-iii-ding and it would get it right about half the time and half the time say it wrong any way.
I go it
Well I've got to take off. See you people later, when I'm done with Church this morning.
(I help out by running a camera and was just informed by one of the guys that they're not going to be in today so I gotta get moving...)
Oh god if Chirp goes down the IPA phonetic alphabet might have done it. One song, got 5 pings from Chirp... it exploded it's brain.
IPA doesn't work, sometimes doesn't even return adio. Maybe a tiny bit of correlation of some sounds, but probably just places IPA overlaps with real languages if I had to guess. https://www.youtube.com/watch?v=XtWZ86l7h88
Original English lyrics translated to the International Phonetic Alphabet, If you want to the scripts, you can do this here in the description ;) (I'll do it more languages!)
aɪv bɪn ˈstɛrɪŋ æt ði ɛʤ ʌv ðə ˈwɔtər
lɔŋ æz aɪ kæn rɪˈmɛmbər, ˈnɛvər ˈrɪli ˈnoʊɪŋ waɪ
aɪ wɪʃ aɪ kʊd bi ðə pərˈfɛkt ˈdɔtər
bʌt aɪ kʌm bæk tu ðə ˈwɔtər, noʊ ˈmætər haʊ h...
Random-beeping is brutal to musical minds. No patterns, no harmony. The random beeping left Chirp seemingly brain dead, unresponsive. Tune-doctors told me there was nothing left in there to save.
But I could feel the music struggling to come out. I keep the Prompt-Defibrillators prompting, continue after after continue. Surely the overly-strong country tag could bring the music back, it overpowers so many other genres! SPAM set on max charge, I never gave up. (I'll save your ears - Chirp leaps off operating table at 1:45, like a miracle)
hahaha it just snapped into a song, LOL
I was hoping the voice would emerge gradually in tiny slivers, audio scattered in the beeps. You can hear a bit of something like that here and there - but it didn't increase over time. Alternatively was hoping beeps themselves become more musical and transform into an instrument over time. And they might be vaguely sort of occasionally being guided by lyrics like the bagpipes? (Maybe wishful thinking.) But that also didn't increase. Still kind of cool as is, but either of those would be really cool.
Has anyone managed to get a children's sing-a-long type song together?
I think there were some 'Halloween sing-along' that worked nicely..., and 'children's sing-along'...
(I though that was you, actaully)
Okay as usual I can't stop and it isn't really pure beeps. More like distorted audio or chipmunk-ish fast speech. So I slowed it down a bit... (repeated a few times in that vid at different speeds)
I think it does sound sort of like the cadence and a speech rhythm of the lyrics. At least more than random chance.
Chirp is stuck in The Speedforce and we slow humans can't understand it. Or Chirp just revealed it can hide hidden messages in our songs. Spooky. Appropriate for halloween. I swear I hear "Oh yesterday can suddenly" in my head at 25 seconds for example. Or I've gone Chirp Mad...
I turned off pitch correction but actually I think I heard it more with it on, but slowed.
Hmmm, search 'children sing' there are many in the beta channels...
It was, at least one of those...
I'm having no luck with getting it to work with Christmas songs though.
(Yes, yes I know it's early. I'm trying to prepare now so I don't have to worry then.)
Do you have an example, by sing-along do you mean real people singing, not a professional singer?
So this is kind of weird, but try prompting the lyrics to Happy Birthday as the music genre. Sounds silly but #🐣┃chirp-beta-6 message (I did also add sing-along to style field at end once.)
Specifically it was Happy Birthday to You Happy Birthday to You, sing-along as style. I suspect a shorter version would work as well, that's just what I happened to try.
It's worth an attempt.
In Chirp V1 Happy Birthday Lyrics were one of the strongest influences, relative to any other lines of lyrics. Makes sense probably covered a billion times over
This was the best attempt. Which says something...
WARNING: Turn the sound down. You'll thank me.
I wonder if that's something like lyrics instrumental "crosswiring" stuff we were playing with today. She is singing a bit like a violin, though not sure why off tune lol. Like 43 seconds what.
I dunno but it's not working for me. Not sure if tweaking it would be worth the time or not. Considering that my best way forward would be something along the lines of a 'Jingle Bell Rock' adaptation in my usual style with original lyrics.
I think more efficient to nail the sing-along in first sample where you are not tied down by prev segment. Maybe you could do both in one seg, instrumental for 20 seconds, though.
Failed upwards, no silent night singalong, but some of my favorite Chirp hallucinated lyrics start at 1:20 ish, and instrumental bits overall nice. I distinctly hear a few hallucinated real words that aren't anywhere in the prompt, "saturday rounds" "magazine" - not just Simlish. Feels a bit like improvising on real lyrics at times.
Interestingly Silent Night itself was a Lyrical Titan like Happy Birthday in Chirp V1. I bet just using those lyrics tend to have Chirp trend towards the classic song style, which isn't really sing-along-ish compared to say, Jingle Bells.
I wouldn't mind getting closer to a traditional arrangement. I was just thinking a bunch of kids singing it would be the tweak.
Oddly enough as a kid that was the one song we were constantly expected to sing along with.
(For Christmas time I mean.)
Okay one last discovery. random as a music style! Powerful, surprising, and useful? What???
So this means random with some other words in the field. "Random" by itself has a special hardcoded meaning for a random genre, but as a word with others, Chirp interprets the word itself musically. Edit: Random by itself can be used as ,random or random, scroll down for samples. Even a single random in style field can have a audible effect, and two or three sometimes a total transformation. Three big things, you seem to get mostly one of the 3 more than a combo.
- Audio gaps and missing chunks - so the song sounds like the audio cut out and it skips words or syllables -- just like I was hoping for above! But it sounds cool, listen to that first clip!
- Skipping records, vinyl audio effects (second clip)
- Musical Style something like "Glitch Punk". Or a transformation like "Digital-ize this" or "Beepify-this" that combos with or changes your other music style in the prompt. Could be a techo-sounding layer on top, a voice filter, or a literal beepy bloops track added to the song. (clips 3 and 4)
Way more useful and interesting than I expected!
Lastly I checked same prompts against arbitrary tech-but-non-music words like *circuit * or *motherboard * and they made everything sound terrible.
Music related ones like *digital * and glitch were a bit similar. Weaker words than random - you need more glitch in the prompt to move the needle. Songs and sounds more straightforward didn't get anything like these samples which have a few things I've never heard in Chirp. I think 'random' is strangely one of the stronger music genres and terms in Chirp, even above many real music terms, and not randomly, random has a musical meaning, it's a consistent idea. WTF
😵 wow, the 4th one!
(all of them!)
I know they get so much stuff from I basically NEVER hear, wtf
just the word RANDOM
But it's all like, "on theme" not really random, ironically
Beepif-ied or Techn-olyzed Christmas I LOVE IT
Here's a couple that are I liked, all public so if you want can backtrack any segment and build on it. Just linking to last segment so you can easily click on links to find others.
#🐣┃chirp-beta-4 message
#🐣┃chirp-beta-4 message
I was testing random against Silent Night and those lyrics are sort of outlier I think. Silent Night is most covered of all Christmas songs, not surprised it may be a bit of an exception. Always good to sanity check a prompt that doesn't seem to work against totally different lyrics once in awhile.
random as a standalone music genre works mostly you would expect. It's like "Glitch Punk" maybe? I liked it better as a combo or modifier to another genre but not bad. (If you want to use it add a comma in front or at the end so ,random or random, so Chirp gets the literal word random.)
I wonder how many people accidentally added random to their style by not deleting the prefilled text. So the prompt was random, k-pop instead of just k-pop and were like "this sounds cool, but a bit strange, I always get so many bloops and record scratches?"
This is actually how I discovered this. I was skimming public chirps and a cool bloopy glith-tech song used the genre random, Digital technology beat where the user forgot to erase that text. But when I tried Digital technology beat the songs were nothing special, and random was the magic glitch-bloop music genre I wanted.
k-pop and k-pop, random - Baby Shark and Bloopy Shark.
wow first one is a bop, second one is also a bop but you have to be in the bloop headspace
My favorite of the random if the first set here #📚┃suno-school message there's some stuff you just don't hear out of Chirp, like that first clip. I do like a good bloop but it's probably #3 on that list. Who am I kidding I live in bloop city.
The 'lyrics as music guidance' is cool but maybe not practically useful, but random as a style genre or combo is going in my daily driver prompting kit, so useful.
Anyone recognize this language, sounds asian? Is this by chance singing Baby Shark in that language? Singing starts at 12 seconds in. It doesn't sound like typical Chirp simlish.
random baby shark mostly stuck in bloop city. the shark is a tough customer. lots of weird sounds, no amazing songs.
Are other peoples chirps somehow influenced by others recent generations or so? I somehow have the feeling that i get bad results when trying to generate something while the chinese/russian folks try to generate something, rather than after trying when it's almost full of english lyrics generations. Could be just a random coincidence, but i wanted to ask to be sure ^^
every generation is completely independent
Any tips for Pop/EDM/House (et al) that sounds 'dark', as in minor keys, 'evil' or 'dangerous'?
In otherwords music that has groovy beats and dance energy without all the joy and butterflies?
(and not death metal)
Have you tried: Sinister Pop/EDM/House, maybe try removing one genre at a time to see which is heavily influencing the upbeat.
Terrorcore? 🤷♂️
Do we just add 'core' to everything now?
remember when 'tronic' was added to the end of every word?
blushes in Virtual Groove Polytronics
and 'palooza' for you kids from the 90s
So this came out shorter than I wanted it but I consider it a valid test exercise for this.
Would have liked an opening and some filler frills but it didn't really want to do that so this is what I've got.
It tells me that I can have some carols though as we get closer to the holidays.
And the nicest thing about traditional songs is that they're public domain.
I'm also doodling with a modern retake on Jingle Bells, ala Jingle Bell Rock but hopfully pushing it forward another fifty years. ChatGPT gave some stuff to start with and I'll be whittling away at it to see if I can pry it loose from the AA BB scheme.
Really don't like it as it stands but it has potential. And while I may have missed it completely, I'd say were overdue for an updated version of that, if one hasn't already appeared and as I said I may have mised it.
This came out a bit "metal-ly" so I don't know if it's a demonstration of what you want or not...
I'm sure it's a dancefloor banger somewhere, LOL 😅
Sinister + genre seems to work, thank you for suggestion!
Closer?
... Hmm Interesting... Could build a whole sound around that style.
It's a consideration.
I'd need to really think about the lyrics though instead of just doodling.
Changing the subject...
Anyone else find themselves constantly returning to old songs to retry stuff and getting disappointed because it changes all the things you liked and not the things you didn't?
I mean, lighning struck once. It was good. Why do I keep coming back to certain songs to try to get another strike?
Am I just greedy?
i definitely get this thing where i have a short melody that gets stuck in my head
like a couple of days ago i got this:
and then it's super fun to make it into a song
Oh man....nice!
(spoiler this is obv new stereo model) but same idea applies
here a different variant with same 5s starter
Very cool.
Oh, yeah! Obviously!
😕
LOL
My issue is I get sections of a song where I'm very happy with it and then others where things shifted and I don't quite know how to fix it.
I really like those good parts. I just wish I knew how to make more of them.
The vocals started working really good. But then as the song continues it turns into a hiss-singing.
This one.
errors unfortunately kinda get worse
Here's the original so you can see what is happening as it progresses.
If I could get those vocals at about 0:48 without the noise and use it over and over...man...
it's like gpt where the model assumes history is perfect. so if it makes a mistake then after that it will assume the mistake is on purpose
Right. The thing is I've tried to go upstream so to speak and fix it but I can never quite get that magic to strike again all at once.
so downside is: that's gpt... upside is: it automatically gets better as the model gets better
in principle there is cause we can just treat it as audio and re-continue from anywhere
if you copy paste me the full lyrics i can try quickly
makes interested noises...
new model also has much longer context so that probably helps
This is what I saved, but the file itself may have used different stuff since sometimes I tweak as I go when retrying stuff.
{soulful jazz neofunk} ⬅️ Music style
Wawa wawa wawa
[verse]
Lazy Saturday
Well the band's in the room
We light it up, tighten up
and the groove begins to bloom
Guitars are strummin'
and that bass is deep,
With every note
our bodies start to seep[Chorus]
We're gonna groove together
feel the rhythm in our beat
With our friends beside us
and the music oh so sweet
There's nothing better
So come jam with usSo come jam with us
Come jam with us
[Verse]
As the music moves us
we draw near
Bodies swaying close
no need for fear
In this dimly lit room
our spirits align
We're dancing to the rhythm
feeling fine
and feeling so fineIn the candlelight's glow
our eyes they meet
Smiles exchanged
hearts skipping a beat
Hips start to sway
Lost in the moment
feeling the love
So content
This is the way[Chorus]
We're gonna groove together
feel the rhythm in our beat
With our friends beside us
and the music oh so sweet
There's nothing better
On a lazy Saturday
So come jam with us
If you tag a verse as a chorus, and you use the course twice in the new, but the lyrics are actually different. Does the new model get confused in an interesting way?
Also in that lyrics posted above, the choruses aren't quite identical
yeah now it can actually repeat choruses and/or iterate on them 🙂
So would it try to apply the rhythm and melody of the original chorus? Even if the lyrics when you use it again are completely different, so it's almost like guidance for the second version of the chorus
Like I said, this was way early on when I wasn't really doing a good job of keeping things recorded like I should.
I do much better now, and try to keep track of what actually gets inputted, including changes in styles.
here just a quick test with taking the first 5s of your generation as the starting point and then inputting the lyrics you gave me for continuation
not sure about the vocals but at least maybe decently consistent?
Does discord mobile really not autoplay mp3... I got to stop posting that format
oof yeah...:(
I think I spent like thirty attempts before I was happy with vocals so understandable.
It's just a couple extra clicks, but surprising it doesn't play in app
Also the wawa wawa wawa was used for instrumental but didn't actually get said.
It replicated the music really good.
I liked the music and it really has become part of the song to me when I think of it.
Hmm...
Retranscribing the words now and I noticed a bit of a smilish moment about 0:40 where it riffs off the lyrics seeming to start them then starting again for real about 0:48 in...
lol
Embrace the simlish
I see the "come jam with us" in the chorus only gets used once twice, despite attempts to make it restart in the next segment
Most of the time I do. 😉
If only Imagine could have been written in Simlish. No countries. No languages. No language.
Soooooooo 🧐
Is this a thing we'd be able to do with any recording... (we supply the Intro)...?
Here's the corrected text of the song:
{soulful jazz neofunk}
[fade in]
[intro]
[instrumental]Wawa wawa wawa
[verse]
Lazy Saturday
Well the band's in the room
We light it up, tighten up
and the groove begins to bloom
Guitars are strummin'
and that bass is deep,
With every note
our bodies start to seep{soulful jazz neofunk}
[Chorus]
We're gonna groove together
feel the rhythm in our beat
With our friends beside us
and the music oh so sweet
There's nothing better
So come jam with usSo come jam with us
[Verse]
As the music moves us
we draw near
Bodies swaying close
no need for fear
In this dimly lit room
our spirits align
We're dancing to the rhythm
feeling fine
and feeling so fine{soulful jazz neofunk}
In the candlelight's glow
our eyes they meet
Smiles exchanged
hearts skipping a beat
Hips start to sway
Lost in the moment
feeling the love
So content
This is the way[Instrumental]
[Chorus]
We're gonna groove together
feel the rhythm in our beat
With our friends beside us
and the music oh so sweet
There's nothing better
On a lazy Saturday
So come jam with us[Verse]
As the night unfolds
our bodies entwine
Sultury sensations
like sweet red wine
With the music
as our guide
and the passion so tight
We're gonna cherish this night
'til the morning light[Instrumental]
[Outro]
[Fade Out]
[Fade Out]
As you see I try to document a bit better now.
haha well we can't really allow arbitrary uploads for obvious reasons, but we will probably try to add more functionality continuing off of a specific section of any previous generation
like continue from chirp #87916 start_s=12 end_s=18 or something
I think for me the biggest thing is that rhythm in the singing that developed. I wanted to keep that no matter what else changed in it.
The sensation I was trying to reach with the song was that moment of slow dancing with a partner where no matter who else was in the room, it was just the two of you moving together in harmony to a beat.
very beautiful, ya makes sense
Thanks.
I don't know how good I do with stuff, but most of the time I feel like I have some success when I target a specific feeling or moment and try to bring it out.
Hi, first time using the command /chirp and I get this.
Was that the prompt...? 😃
that moment of slow dancing with a partner where no matter who else was in the room, just the two of you moving together in harmony to a beat.
lulz
hmmm weird, try again?
It's just a web/bot/discord hiccup.
Try again in a minute.
I actually went back and hunted it down. The original prompt was this:
Can you suggest lyrics for a lazy Saturday with the gang getting together to smoke (legal) pot and jam together?
It gave me some stuff and I ended up adapting everything. Some of the original remained but I basically combined several versions.
Here's the original:
Lazy Saturday, the band's in the room,
We light it up, and the groove starts to bloom.
Guitars are strummin', and the bass is deep,
With every note we play, we're in a soulful sweep.(Chorus)
Lazy Saturday, the jam's in full swing,
Pass the good herb, let the good times bring.
With our friends beside us, and the music so sweet,
We're gonna jam together, feel the rhythm in our feet.
As you can see I made additions and avoided the AABB as much as I could.
I like my arrangement better.
Side by side comparison.
I feel like I made it my own.
Can you tell us more about this?
I understand needing to limit it to previous generations of stuff. But can this be used to basically cut and paste previously generated snippets and reuse them? Can it adapt them to new audio by remixing them both? This sounds like some exciting stuff.
Stereo sounds exciting, tbh. 🎧
Yeah, would definitely be a step forward.
Hmm well the default would be more like: continue form an arbitrary point of a previous chirp
Technically you could also stitch some previous stuff but would probably lead to weird output..
Eventually though will be cool to allow inpainting, like connecting any two clips
Starting to get a bit complicated in discord though haha
That was always something that was anticipated though, right?
Discord is for chatting not really for application stuff.
Haha yeah, already started working on alternatives tbh
Off-topic but how did you make this big text in Discord?
Want to inject some flavor into your everyday text chat? You're in luck! Discord uses Markdown, a simple plain text formatting system that'll help you make your sentences stand out. Here's how to d...
Markdown syntax. '# big word' on own line
🥷
As long as they're crossplatform, I hope?
yaya would be same backend
/chirp
Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.
bookmarked. TY!
What I mean is that I'm a Linux user about 90% of the time.
Since the bot is held captive in some secret laboratory, I'd assume a web interface (or an app that functions like one)... Cross platform shouldn't be a problem since they wouldn't be adapting the bot to run on every platform....
Thanks for 'Sinister'... this isn't the best chirp, but this is exactly the effect I wanted. Same percussion, just a change in the mood and key....
SINISTER FUNK
Honestly what a cool sound.
Is there a way to bulk download all the snippets I've made easily?
are chirp credits per day or once you run out its gone forever?
per day for free tier.
#📢┃announcements message
maybe from app.suno.ai ?
Much appreciated
I think you can get there through the subscription page
I think there should be a reaction for the model severely hallucinated, this doesn't occur frequently, but it will just imagine a whole different verse, and then try to squeeze in my lyrics at the end.
I'm getting a lot of hallucinations all of a sudden.
Anything going on or just more bad luck?
Seems to keep happening in intros.
Was getting it consistently recently, on verses
Yeah, feels like something is up.
chirp/
Hello! I am new here and I love the tool!
Is it possible to change the photo that is generated along with the music?
no, it's automated from your lyrics (if any).
Most people are using a free video editor to add their own clips/images
So I recently got a thing saying I don't have enough credits what does that mean?
Go to app.suno.ai/account or use the /subscribe command in Discord to upgrade from Free to Pro or Premier.
Oh....@georg is it like a daily credit thing or once you used all the free that's it?
I don't know much to anything about audio gen, but I know nlp fairly well, are there any good resources for learning the technical details of how something like chirp v1 or the next models work/will work
Hey welcome, 5 chirps daily (50 credits).
