Why is there no sound sometimes when using the 5_HP-Karaoke-UVR.pth model?

[Lixi20]

I found that it may be related to the parameters, please help me

from pathlib import Path

from audio_separator.separator import Separator

def main():
    input_audio = "/home/geek/workspace/TestSample/Tom_Holland.wav"
    output_dir = "/home/geek/workspace/TestSample/"
    model_file_dir = "/home/geek/.cache/audio-separator-models"
    vocal_separation_model_filename = "UVR-MDX-NET-Voc_FT.onnx"
    de_reverb_model_filename = "5_HP-Karaoke-UVR.pth"
    output_format = "wav"

    vocal_separation_separator = Separator(
        model_file_dir=model_file_dir,
        output_format=output_format,
        output_dir=output_dir,
        output_single_stem="Vocals",
        sample_rate=44100,
        mdx_params={"hop_length": 1024, "segment_size": 256, "overlap": 0.25, "batch_size": 16, "enable_denoise": False},
        vr_params={"batch_size": 16, "window_size": 320, "aggression": 10, "enable_tta": True,
                   "enable_post_process": False, "post_process_threshold": 0.2, "high_end_process": False},
    )

    vocal_separation_separator.load_model(model_filename=vocal_separation_model_filename)

    vocal_separation_output_file_path = vocal_separation_separator.separate(input_audio)[0]
    print(vocal_separation_output_file_path)

    vocal_separation_separator.load_model(model_filename=de_reverb_model_filename)

    de_reverb_output_file_path = \
    vocal_separation_separator.separate(Path(output_dir) / vocal_separation_output_file_path)[0]
    print(de_reverb_output_file_path)


if __name__ == "__main__":
    main()

When I set the parameters as follows：
mdx_params={"hop_length": 1024, "segment_size": 256, "overlap": 0.25, "batch_size": 16, "enable_denoise": False}, vr_params={"batch_size": 16, "window_size": 320, "aggression": 10, "enable_tta": False, "enable_post_process": False, "post_process_threshold": 0.2, "high_end_process": False},
the de_reverb_output_file_path is no sound：
https://github.com/Lixi20/audio_separator_test/blob/main/TestSample/enable_tta_false/Tom_Holland_(Vocals)UVR-MDX-NET-Voc_FT(Vocals)_5_HP-Karaoke-UVR.wav

And then I set the parameters as follows：
mdx_params={"hop_length": 1024, "segment_size": 256, "overlap": 0.25, "batch_size": 16, "enable_denoise": False}, vr_params={"batch_size": 16, "window_size": 320, "aggression": 10, "enable_tta": True, "enable_post_process": False, "post_process_threshold": 0.2, "high_end_process": False},

the de_reverb_output_file_path has sound：
https://github.com/Lixi20/audio_separator_test/blob/main/TestSample/enable_tta_true/Tom_Holland_(Vocals)UVR-MDX-NET-Voc_FT(Vocals)_5_HP-Karaoke-UVR.wav

I didn't set the enable_tta parameter before, and there was sound, but now it doesn't work. Why is this?

[Lixi20]

@beveradb

[beveradb]

Sorry @Lixi20 but I don't know, I didn't train that model or write the original implementation for the VR architecture!

If you really want to figure it out, you can investigate what the code is doing and try to debug it:

• https://github.com/nomadkaraoke/python-audio-separator/blob/main/audio_separator/separator/architectures/vr_separator.py#L279
• https://github.com/nomadkaraoke/python-audio-separator/blob/main/audio_separator/separator/architectures/vr_separator.py#L326

Good luck! If you find a fix, please raise a PR and I'll test & merge it :)

[Lixi20]

here

here

code

Hope it helps you! !! !!

@beveradb

[beveradb]

Thank you so much @Lixi20 - this fix is now live in audio-separator version 0.19.2 🎉