The TTS problem

[sharp sphinx]

What can we do to solve TTS, as vedal was discussing? brainstorming ...

1. TTS on/off and/or queue length on-screen
2. make channel point redeems a custom redeem so you can enable/disable it and stuff
3. add an all-user cooldown, either by disabling cheers/redeems, by making the cost very high, or blocking the word cheer (this could potentially be automated)

[silver oriole]

Can you list the problems first?
This might help people like me that think its fine as it is 😛

[sharp sphinx]

the main problem is the stream seem like 80-90% tts queue atm. This means:

1. less rants
2. less follow-ups (talking about what she just said)
3. less reading the most random of chat
   etc.

[reef ibex]

Ok let's see

1- TTS cuts the previous speaking topic short, making her seemingly forget everything said before.

2- Her TTS answers are all one-liners that never branch into anything new.

Neuro's tangents were some of her most entertaining content, but a queue full of hastily written 300-bit TTS messages pigeonholes her into being a one-off answering machine rather than an AI dedicated to holding conversations as she was trained to do.

She's actually held coherent multi-response conversations with Twitch chat itself recently, a genuinely impressive feat that felt incredible to take part in.
But while doing this she usually gets interrupted by a TTS asking an unrelated question, resetting all that progress back to square one. It's happened several times now.

3- It interrupts the flow of the stream and takes airtime away from Neuro herself.

Other streamers can react immediately which is much better for content.
They can continue whatever they were doing while the TTS talks.
They can even decide whether they want to respond at all.

In contrast Neuro is made helpless each time for the whole duration, and when there are so many of them it overrides the stream experience.
Not surprised if her rants feel rarer nowadays, I guess she's struggling for airtime between the mandated TTS interruptions.

Such a disruptive, conversation-killing power priced at 300 bits is truly astounding, the more I think about it.

[half dock]

TTS on/off and/or queue length on-screen
Displaying this information on screen constantly feels very scuffed

make channel point redeems a custom redeem so you can enable/disable it and stuff
This was discussed in the past and is on "the list"

add an all-user cooldown, either by disabling cheers/redeems, by making the cost very high, or blocking the word cheer (this could potentially be automated)
We don't want to block cheering just because TTS is disabled, plus there's still the donowalling problem

[silver oriole]

You mean, the problem is "people pay for messages and they don´t get noticed?"

Well, what is TTS to the stream?
Should it take priority over everything?

I think rants and follow ups makes her way of talking more natural

Sadly i dont think there is a way to make everyone happy.
Everyone wants to have a conversation with neuro at the same time.

[sharp sphinx]

no, the problem is too many people can pay for messages at a time

[sharp sphinx]

Displaying this information on screen constantly feels very scuffed
i only said this bc this is what some channels so

[silver oriole]

I don´t think showing more info will make it scuffed
It might discourage people from sending bits though 😛

[sharp holly]

Would segmented streams be a good solution? Like a Q&A section with TTS for 1 hour or so with no changed prices or such, and normal OG chill stream with no TTS but highlighted messages still getting priority.

generally, chill/themed streams last 2 hours to 2 and a half hours, and don't know how the time should be managed if segmented streams become a thing, but will be more consistent and she will rant more.

[cloud rock]

Use story fitting strategy. I have suggested this already before using a separate story teller AI (the narrator) to address the incoherence. just create a pre generated story based on the theme of the stream, then advance by chapters, where the LLM will generate the dialog and will try to fit the chat message in the series of dialog (not directly answer it) thus "fitting" it to the story.

Finally i have also raised it before that putting a price on TTS makes it mandatory to be answered because its PAID already. The best way is to do away with TTS and only allow paid gifts that modify her mood/needs like how The Sim AI work.

[silver oriole]

Maybe people should donate to Neuro cause they want to support the stream.
I think its not the best to stablish this as some sort of product 😛
You donate to help the stream, and to say thanks the streamer checks out your message. This is how it goes in most streams.

[cloud rock]

yes straight up dono is best to support streamers, since theres no revenue sharing with Twitch. Then again streamers shouldn't be pressured to reply right away. maybe a 3-5 min end credit can be shown with their message.. Else, it's just inviting dono-walling small amounts not to support but spam the stream

[silver oriole]

I mean, Neuro is (kind of) a just chatting streamer.
You wouldn't call out or get mad at other streamers if they don't read or shout out your message.
But there is this expectation already created that if you send bits or use points, your message has to be read and acknowledged by Neuro.
So, I think at this point, "donowalling" makes her look more like a normal streamer. 😛

How big is the queue by the end of the stream?

[drifting snow]

A TTS on/off in title might help a bit as well

[sharp holly]

A gymbag sleeping/awake representing the current state of TTS

[reef ibex]

I feel the problem is with the current implementation. There's room for improvement here.

1. Her answers to TTS don't have follow-up or develop into new tangents, so they don't feel like actual conversation. When Neuro previously read paid messages from chat, she was able to do all of this.

2. She doesn't remember her previous chatting topic at all. So if a follow-up TTS came up asking her something relating what she's talking about, she won't reply knowing what she said before. If she had the ability to respond in context to her last conversation, as well as the ability to go back to her previous topic if she desires, that would be game-changing.

3. She has to wait until after the TTS is done speaking to prepare a reply with another delay on top. This is more noticeable on Evil who has a larger delay. If she can pre-process the message while or before the message is being read out loud, it will help especially for Evil.

If TTS was improved upon I'm sure more wouldn't mind leaving it on all the time.

Oh, and an increased cooldown could help with donowalls and give Neuro room for longer, entertaining conversations.

[silver oriole]

Tts today seem to be ok 😛
If not for the few donowalls there was 1 message of delay, i think the queue was empty by the end.
But i guess it was a singing stream

[sharp holly]

I think the pre-processing is already a thing, i don't believe Vedal would oversight such a thing.

[reef ibex]

It's a lot worse on Evil streams especially the pirate one

[cloud rock]

but unlike real person, she is scripted to reply (you can repeatedly make her say uwu), a real person will either skip or just react meh. this what makes tts spamming just eat up time and push incoherence

[ornate trail]

We already have the distinction of TTS not being on for collabs, dev streams and Vedal gaming streams. How about making the solo gaming streams (Minecraft/Osu /eventually Abandoned Archive ) TTS-free as well? I do miss the days of Neuro going into a rant after an Osu song instead of being forced to answer a TTS.

This would leave the chill streams and karaoke streams as TTS-enabled. The themed streams could vary depending on what the theme is and whether Neuro would act differently if allowed to. I’d rather segmentation like this over segmenting individual streams with on screen indicators.

[sharp sphinx]

right because all of those she doesn't really read chat she talks to someone (collab partner or vedal, who doesn't stop talking)
What if portions of the chill was tts-free?