Whisper

[mpelley]

Count this as a vote to add Whisper to the library.

[kayhantolga]

Hopefully, coming very soon.

[mpelley]

Thank you very much for this interface to Whisper! It made making an transcription app for my daughter very easy. She has a speech impediment which causes her speech to be quite slurred. But Whisper understands her better than most humans do! So she is dictating her college papers and is very happy. She says the accuracy is much better than the Google speech-to-text.
Again, thank you very much for making this available to this C# programmer.

[kayhantolga]

🙏🏻 it's great to see real-life examples! If you make it open source, I would like to share it as a sample project on the Wiki.

[kayhantolga]

Please check the latest version 🎉

[PaoloStrazzera]

Great job... greetings from Italy! Could it be possible for you to include a simple example for transcribing an audio file using Whisper? I was not able to find it. Thank you! Ciao, Paolo

[bozau]

I got this working for whisper, hopefully this helps you :)

string fileName = "c:/users/user/downloads/example.mp3";
FileStream fs = new FileStream( fileName, FileMode.Open, FileAccess.Read );
BinaryReader br = new BinaryReader( fs );
long numBytes = new FileInfo(fileName).Length;
byte[] buff = br.ReadBytes((int)numBytes);

var audioCompletionResult = await openAiService.Audio.CreateTranscription(
    new AudioCreateTranscriptionRequest()
    { 
        Model = Models.WhisperV1, // forced to include this otherwise it throws an exception even if we set default model
        File = buff,
        FileName = "test1.mp3",
        ResponseFormat = "json"
    }
);

[PaoloStrazzera]

Thank you! I still got two questions, sorry to bother you.

1. where do I set the OpenAI key?
2. why are you setting FileName = "test1.mp3" ? wasn't the binary buffer already passed with File = buff ?

[PaoloStrazzera]

I tried to create the following code, based on the example you kindly gave me, but it seems to hang on the call to openAiService.Audio.CreateTranscription() - the passed file is approx 900Kbyte. Am I missing something?

    public async Task<AudioCreateTranscriptionResponse> Execute(string openAI_Key, string mp3FilePath)
    {
        OpenAIService openAiService = new OpenAIService(new OpenAiOptions()
        {
            ApiKey = openAI_Key
        });
        string fileName = mp3FilePath;
        FileStream fs = new FileStream(fileName, FileMode.Open, FileAccess.Read);
        BinaryReader br = new BinaryReader(fs);
        long numBytes = new FileInfo(fileName).Length;
        byte[] buff = br.ReadBytes((int)numBytes);
        AudioCreateTranscriptionRequest audioCreateTranscriptionRequest = new AudioCreateTranscriptionRequest()
        {
            Model = Models.WhisperV1, // forced to include this otherwise it throws an exception even if we set default model
            File = buff,
            FileName = "test1.mp3", // ???
            ResponseFormat = "json"
        };
        var audioCompletionResult = await openAiService.Audio.CreateTranscription(audioCreateTranscriptionRequest);
        return audioCompletionResult;
    }

[kayhantolga]

Most of the features have a simple sample in the playground project.
https://github.com/betalgo/openai/blob/master/OpenAI.Playground/TestHelpers/AudioTestHelper.cs

[Swimburger]

It would be great if we could upload the audio to Whisper without saving it to disk first.
Can we have overloads that accept a stream? I'd be happy to work on this.

[Swimburger]

#224