search

saharmor / whisper-playground

Build real time speech2text web apps using OpenAI's Whisper https://openai.com/blog/whisper/
MIT License
777 stars 140 forks source link

USING THIS SETUP ON CPU #42

Open shanky100 opened 1 year ago

shanky100 commented 1 year ago

Hi, just wanted to ask if it is possible to use this repo using CPU , if yes then do I need to make any specific changes to the code or any other additional step. OS: Ubuntu 20 LTS and python 3.8

shanky100 commented 1 year ago

I managed to get the system up and running, whatever I speak the analog signal changes but I do not received any transcription on the UI/web page. I cna see the transcription being generated in the terminal logs.

Is someone else facing the same issue, any suggestions will be helpful.

saharmor commented 1 year ago

Can you attach the console log (a screenshot would do) + a screen recording?

ethanzrd commented 1 year ago

Following Sahar's request, please make sure that the screen recording is at least 15 seconds if you're using the real-time mode with a transcription timeout of 5 seconds :)

shanky100 commented 1 year ago

@saharmor Thank you for the reply, I am attaching a screen recording which you can check, although in terminal I see the transcriptions after some time but I do not get nay print on the web page UI.

https://github.com/saharmor/whisper-playground/assets/132049696/f3dd5e41-c30e-45af-b343-32c49ef5e8c2

image

I hope this helps to get a better understanding on the challenge I am facing.

ethanzrd commented 1 year ago

That's odd. I don't know what could be causing this. I have an odd request, let me know if you want me to clarify.

Once you see "Transcription generated", forcefully terminate the program (Ctrl +C), go to the project files, open the backend folder, and there you should see a folder called temp. Within that folder, there's a file called batch.wav, could you please upload it to Drive and share the link? Or compress it and drag-and-drop it?

shanky100 commented 1 year ago

https://github.com/saharmor/whisper-playground/assets/132049696/d4355b4c-8e27-422f-b345-b76e69bf1f87 @ethanzrd I just checked what you suggested but I don't see any temp folder getting created in the backed folder.

I by mistake said backed folder in the video but I mean temp folder only. YOu can check the recording

Also, I forcefully terminated the program after a transcript got generated for a segment, but still I cannot locate any temp folder being created in the backed folder or any batch.wav file.

ethanzrd commented 1 year ago

You're right, I was mistaken. This only applies to the sequential mode. Speaking of which, could you try it and let me know if you're having any problems there? If you are, then the batch.wav file will be present for you to include :)

shanky100 commented 1 year ago

@ethanzrd I checked with sequential mode. There it generated the wav file and it is having the voice recording. But I neither see any transcription in my terminal logs nor on the whisper playground WEB PAGE.

I am sharing the wav file. temp.zip

image

ethanzrd commented 1 year ago

Huh, that is odd. I'll test a bit with the recording tomorrow :)

shanky100 commented 1 year ago

One more thing I checked , I am getting the transcript in my

tag but the issue is that it is not getting displayed on the web page. Here is a screenshot for the same. Any guess/suggestion for this behaviour?

image

shanky100 commented 1 year ago

@saharmor Thank you for the reply, I am attaching a screen recording which you can check, although in terminal I see the transcriptions after some time but I do not get nay print on the web page UI.

issue.mp4 image

I hope this helps to get a better understanding on the challenge I am facing.

is there any update/solution which can be recommended?

ethanzrd commented 1 year ago

As for the HTML problem, have you tried expanding the window?

I don't know what could be causing your transcription issue as your audio is valid, I'll try to run it on a clean Windows installation and get back to you.

shanky100 commented 1 year ago

I made a fresh installation and had to make change in package.json "start": "react-scripts --openssl-legacy-provider start" to "start": "react-scripts start --openssl-legacy-provider" Also I updated my chrome to latest version.

I can see the transcription now, but in starting for most of the segments the transcription being generated is empy, after 30 secs it starts to transcribe something but before that I get no transcript. Is there any suggestion/solution which can be recommended.

image

ethanzrd commented 1 year ago

That's what I couldn't understand. I'm currently a bit busy with other things, but once some time frees up, I'll run a fresh Windows installation and run the project to see if there's any dependencies we might have missed that could be causing this strange behavior.