Transcribe and translate audio/videos files to text

[crude mantle]

Hey there. Sharing a project I built myself using Whisper and GPT AI, in it you can:

• Transcript audio/videos to text
• Translate to another languages
• Export all to PDF

Website: https://audioscript.aurelian.dev/

[grizzled pecan]

Trying it now

Wow impressive!

[grizzled pecan]

in some instances however...I was a bit disappointed with the results. One item I had encoded from my native language returned a transcription in Russian

[sullen osprey]

Is there some sort of limit for the length of the videos?

Just presents a loading circle and then stops after >30 seconds

[crude mantle]

@sullen osprey What's the size of your video? mb and minutes

[sullen osprey]

@crude mantle quite substantially large since they are mp4s (150MB to 70MB & 40mins to 50mins)

I could convert/split the mp4s to just an audio file I suppose, but knowing the limits would be helpful

[crude mantle]

@sullen osprey I'll upload a FIX to find out the exact problem, when I finish I'll let you know here. But you can convert your mp4 to mp3, I think that would work

[sullen osprey]

any suspected issues/problems you're thinking about?

@crude mantle converted to mp3 and it does the same (40MB)

[crude mantle]

I thinks its about the size. Maybe the max size is about 25MB

[sullen osprey]

Would there be a fix to that? Or should I just split the audios?

[crude mantle]

Can you try to split the audios?

[sullen osprey]

this works swimmigly 👌 was just curious :)

[crude mantle]

Yeah, I think the max size is about 25mb. I'll try to implement a feature that decreases the file size

[sullen osprey]

@crude mantle does that jeopardize the quality of the transcription? I'd say as long as it says 25MB limit you'd be good 👌

[crude mantle]

@sullen osprey I don't know if it will compromise, but I will run some tests.

[uncut bolt]

Good work @crude mantle .can I sent a url link of TikTok video , and get the transcription ?

[crude mantle]

@uncut bolt You can download the tiktok video and insert the video to the app to get the transcription

[sullen osprey]

@crude mantle shot you a DM!

[uncut bolt]

Any other inputs anyone has seen which can take url link (internally does this download) and does convert to text..

Downloading videos takes time and also where do u store. Bit of management right.

[crude mantle]

@uncut bolt I don't make any download in the app.

In my github I have a python script that do the same but with youtube urls. I'm sure that I can do the same with tiktok.
Basically, you insert any type of youtube URL and it will do the transcription for you.

[uncut bolt]

Yes @crude mantle I tried with YouTube url. Some other open source repo. It can be done with any url I think. Let me know if it works. How does this costing work? Whisper is still open source right, so is it free?

[crude mantle]

@uncut bolt OpenAI has $18 of trial usage. I already spent $3.95

[uncut bolt]

Whisper is open sourced right. Are they charging it? @crude mantle

[crude mantle]

@uncut bolt Yes, it is $0.006 / minute

[grizzled pecan]

#1080675368303865967 but it seems to not work on my end

also...the demo site is down

[warm crypt]

The docs say 25mb max filesize, but I lose API connection for anything above 20mb

Splitting something into more parts doesn't increase the total amount of tokens. At least not by more than a few 100

So I just split it to 10 minute audio-segments now

Now I'm working on a project to insert the timestamps and speakernames from google meet subtitles into a whisper transcript. (Or vice versa)

Whisper transcript is stellar, but without timestamps and speakernames, it feels a bit inadequate

I think I should probably try to build a chrome extension to tap directly into the google meet session somehow

[crude mantle]

@warm crypt I'm working on it. People will allow to use 100MB + as files

[grizzled pecan]

When will the site come back?

[crude mantle]

@grizzled pecan Next week.

@grizzled pecan Were you using?

[grizzled pecan]

Yes

[crude mantle]

@grizzled pecan @uncut bolt Guys, I launch the app again: https://audioscript.aurelian.dev

@warm crypt I upgraded the app with a PRO version, now you can upload file up to 500mb

[severe flare]

Is there any way to make this work in real-time without an upload or file based process?

[grizzled pecan]

As far as I remember it requires a file upload. Do you mean you want AI to immediately interpret the audio you speak, not upload?