bigvganinference 0.0.3

An unofficial minimal package for using BigVGAN at inference time

BigVGAN Inference

An unofficial minimal package for using BigVGAN at inference time

Installation

pip install bigvganinference

or install from source:

git clone https://github.com/thunn/BigVGANInference.git
cd BigVGANInference
poetry install

Usage

Loading model is as simple as:

from bigvganinference import BigVGANInference

# -- model loading ---

# model is loaded, set to eval and weight norm is removed
model = BigVGANInference.from_pretrained(
    'nvidia/BigVGAN-V2-44KHZ-128BAND-512X', use_cuda_kernel=False
)

# also supports loading from local directory
model = BigVGANInference.from_pretrained(
    "path/to/local/model", use_cuda_kernel=False
)

# --- usage example ---

path_to_audio = "path/to/audio.wav"
wav, sr = librosa.load(path_to_audio, sr=model.h.sampling_rate, mono=True)

# get mel spectrogram using bigvgan's implementation
mel = model.get_mel_spectrogram(wav)

# generate waveform from mel
# note: torch.inference_mode() is used internally
output_audio = model(input_mel)

See the example for full usage example.

Acknowledgements

This is an unofficial implementation based on original BigVGAN repository.

License

This project is licensed under the MIT License. See the LICENSE file for details.