LVC

Unofficial pip-installable Python package for LVC-VC (zero-shot voice conversion) by Wonjune Kang, Mark Hasegawa-Johnson, Deb Roy.

NOTE: I am not affiliated with the authors of LVC-VC (the individuals listed above).

Installation

pip install lvc

Usage

You can either convert files or convert arrays.

Convert Files

from lvc import LVC, LVCAudio
l = LVC()

l.infer_file(
    'orig.wav',
    'sample.wav',
    'target.wav',
)

Convert Arrays

from lvc import LVC, LVCAudio
import librosa
import soundfile as sf
l = LVC()

o_y, o_sr = librosa.load('orig.wav')
s_y, s_sr = librosa.load('sample.wav')

o, sr = l.infer_array(
    LVCAudio(o_y, o_sr),
    LVCAudio(s_y, s_sr),
)
sf.write('out.wav', o, sr)

Don't use XL model

You can optionally use a smaller model for lower-quality but faster results:

l = LVC(use_xl_model=False) # default True

Debugging

If you get an error it's because the model only currently supports short clips. I'm planning to fix this soon, but PRs are welcome!

License

MIT