How to use "initial prompt" wit stable-ts

[cetiny]

First of all, thank you for sharing this great code. I have been testing it for a few days for my project. For one of my transcriptions, I need to use an initial prompt due to missing punctuation.

However, I am unable to use "initial_prompt" argument for the transcription task when I use stable-ts. This works however with the original model. I was able to use "prompt" instead with the same prompt text, but the result is not the same.

This doesn't work

def transcribeWhisper():
    args = {
        "verbose": False,
        "task": "transcribe",
        "language": LANGUAGE,
        "initial_prompt" : initial_prompt,
        "best_of": 5,
        "beam_size": 5,
    }

    model = whisper.load_model(MODEL, device = DEVICE)
    modify_model(model)
    result = model.transcribe("shifted_audio.wav", **args,)
    return result

This works but the result is not ideal (I keep missing the punctuation)

def transcribeWhisper():
    args = {
        "verbose": False,
        "task": "transcribe",
        "language": LANGUAGE,
        "prompt" : initial_prompt,
        "best_of": 5,
        "beam_size": 5,
    }

    model = whisper.load_model(MODEL, device = DEVICE)
    modify_model(model)
    result = model.transcribe("shifted_audio.wav", **args,)
    return result

This works (original model)

def transcribeWhisper():
    args = {
        "verbose": False,
        "task": "transcribe",
        "language": LANGUAGE,
        "initial_prompt" : initial_prompt,
        "best_of": 5,
        "beam_size": 5,
    }

    model = whisper.load_model(MODEL, device = DEVICE)
    # modify_model(model)
    result = model.transcribe("shifted_audio.wav", **args,)
    return result

[jianfch]

Should be working now in fecb375.

results = model.transcribe(..., initial_prompt='a prompt')

[cetiny]

Great! It is working now. Thank you for the quick fix.