Predict using code

[Talhavaival1]

Predict using code

[Talhavaival1]

I wanted to know what should I put in wav, sr = translator.synthesize_speech(<speech_units>, <tgt_lang>) , speech units part

[Mortimerp9]

your sample is to synthesize speech from the output of the translation step. Check https://github.com/facebookresearch/seamless_communication/blob/main/scripts/m4t/predict/predict.py as this has an example of predicting with code.

[Mortimerp9]

https://github.com/facebookresearch/seamless_communication/blob/main/src/seamless_communication/models/inference/translator.py#L150 has more details

[l4b4r4b4b4]

yeah I am stumbling over tihs one as well. And when I plug the tensor response from predict as follows, it does generate a small 2 sec audio, but without any content. But I just started playing around 2 hours ago ...

translated_text, wav, sr = translator.predict(
        content, mode, target_lang, src_lang=source_lang
    )

    # Save the translated audio generation.   
    # wav, sr = translator.synthesize_speech(wav, "eng")
    unique_filename = str(uuid.uuid4())  # Generate a UUID and convert it to a string
    file_extension = ".wav"  # Replace with the desired file extension
    file_path = os.path.join(folder_path, unique_filename + file_extension)

    # # Save the translated audio generation.
    torchaudio.save(
        file_path,
        wav[0].cpu(),
        # TODO test gpu
        # wav[0].cpu(),
        sample_rate=sr,
    )