|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- model_hub_mixin |
|
|
- pytorch_model_hub_mixin |
|
|
base_model: |
|
|
- Alethia/BigCodec |
|
|
pipeline_tag: feature-extraction |
|
|
--- |
|
|
|
|
|
# BigCodec |
|
|
|
|
|
Python library for bigcodec. This is something that the author of bigcodec didn't do, so I'm doing it for him. |
|
|
|
|
|
## Setup |
|
|
|
|
|
``` |
|
|
pip install bigcodec |
|
|
``` |
|
|
|
|
|
## Usage |
|
|
|
|
|
``` |
|
|
import torch.accelerator |
|
|
from bigcodec import BigCodec |
|
|
import torchaudio |
|
|
|
|
|
codec = BigCodec.from_pretrained("intexcp/bigcodec") |
|
|
|
|
|
wav = torchaudio.load("enc.wav")[0] |
|
|
wav = wav.unsqueeze(0) |
|
|
device = torch.accelerator.current_accelerator().type if torch.accelerator.is_available() else "cpu" |
|
|
codec = codec.to(device) |
|
|
wav = wav.to(device) |
|
|
with torch.no_grad(): |
|
|
enc = codec.encode(wav) |
|
|
dec = codec.decode(enc) |
|
|
print(enc) |
|
|
torchaudio.save("dec.wav", dec.squeeze(0).cpu(), 16000, encoding="PCM_F") |
|
|
``` |
