supershaneski / openai-whisper-api Goto Github PK

It seems to work on my desktop PC and android devices but as soon as I try it on a iOS devices, the server gets this error from OpenAI:

{
  error: {
    message: "Invalid file format. Supported formats: ['flac', 'm4a', 'mp3', 'mp4', 'mpeg', 'mpga', 'oga', 'ogg', 'wav', 'webm']",
    type: 'invalid_request_error',
    param: null,
    code: null
  }
}

The file is created successfully in the directory and I can play it with a media player.

After I record with this web application, I get this error in the console ( I made sure my API key was the right one )

[Log] NotSupportedError: mimeType is not supported
MediaRecorder
handleStream

[Log] handle stream2...
[Warning] The resource http://localhost:3005/_next/static/chunks/polyfills.js was preloaded using link preload but not used within a few seconds from the window's load event. Please make sure it wasn't preloaded for nothing.
[Log] NotSupportedError: mimeType is not supported
[Log] handle stream2...
[Log] [send data] – "3:31:14 PM"
[Error] Failed to load resource: the server responded with a status of 500 (Internal Server Error) (api, line 0)
[Log] SyntaxError: The string did not match the expected pattern.
(anonymous function)

Your project is wonderful, and I was able to run the project on local host. When I moved to Heroku, I received 503, service unavailable error, when mainPage.js attempts const response = away (fetch(url.... I see console.log [send data], but [received data] is not reached.

I thought the issue might be related to .pem for SSL, but I understand that Heroku doesn't require that element. Any guidance would be greatly appreciated.

Please provide what is the procedure for making this run from 0 in layman terms.

I have tried your nice project. But I find it cost a very long to get the result. I have used Azure speech recognizer, it is about 4-5 to get the result. But I think whisper cost about 10 seconds. It is very hard to use it in the production.

Hello,
Thank you for the hard work and effort you put in sharing this with excellent documentation.
I have successfully managed to install and run it with API on localhost.
But i would like to deploy it to Vercel (which most NextJs developers will want to use), and while the build successfully deploys and the page loads, it returns an error when i try to transcribe:

- error Error: ENOENT: no such file or directory, open 'public/uploads/file168369759598641389.webm'

I suppose this is related to how NextJs deals with the public folder path (as a source for copying files during build, not as an actual subfolder, so theoretically the correct path for the file on a live server should be 'uploads/file168369759598641389.webm' - i think!

Or is this relevant: https://vercel.com/guides/how-to-upload-and-store-files-with-vercel ? Perhaps it's something else and the files should be uploaded to a CDN or S3? that would be an amazing feature to have.