README.md

---
license: apache-2.0
tags:
- audio
- speech
- audio-to-audio
- speech-language-models
datasets:
- amphion/Emilia-Dataset
- facebook/multilingual_librispeech
- CSTR-Edinburgh/vctk
- google/fleurs
- mozilla-foundation/common_voice_13_0
- mythicinfinity/libritts_r
---

# Model Details

Distill-NeuCodec is a version of NeuCodec with a compatible, distilled encoder. 

The distilled encoder is 10x smaller in parameter count and uses ~7.5x less MACs at inference time. 

The distilled model makes the following adjustments to the model:
* Swap the notoriuously slow [BigCodec](https://arxiv.org/abs/2409.05377) acoustic encoder for the [SQCodec](https://arxiv.org/abs/2504.04949) acoustic encoder (70m → 36m)
* Swap the [w2v-bert-2.0](https://huggingface.co/facebook/w2v-bert-2.0) semantic encoder for [DistilHuBERT](https://huggingface.co/ntu-spml/distilhubert) (600m → 21m)

Our work is largely based on extending the work of [X-Codec2.0](https://huggingface.co/HKUSTAudio/xcodec2) and [SQCodec](https://arxiv.org/abs/2504.04949).

- **Developed by:** Neuphonic
- **Model type:** Neural Audio Codec
- **License:** apache-2.0
- **Repository:** https://github.com/neuphonic/neucodec
- **Paper:** [arXiv](https://arxiv.org/abs/2509.09550)
- **Pre-encoded Datasets:**
  - [Emilia-YODAS-EN](https://huggingface.co/datasets/neuphonic/emilia-yodas-english-neucodec)
  - *More coming soon!*


## Get Started

Use the code below to get started with the model.

To install from pypi in a dedicated environment, using Python 3.10 or above:

```bash
conda create -n neucodec python=3.10
conda activate neucodec
pip install neucodec
```
Then, to use in python:

```python
import librosa
import torch
import torchaudio
from torchaudio import transforms as T
from neucodec import DistillNeuCodec
 
model = DistillNeuCodec.from_pretrained("neuphonic/distill-neucodec")
model.eval().cuda()   
 
y, sr = torchaudio.load(librosa.ex("libri1"))
if sr != 16_000:
    y = T.Resample(sr, 16_000)(y)[None, ...] # (B, 1, T_16)

with torch.no_grad():
    fsq_codes = model.encode_code(y)
    # fsq_codes = model.encode_code(librosa.ex("libri1")) # or directly pass your filepath!
    print(f"Codes shape: {fsq_codes.shape}")  
    recon = model.decode_code(fsq_codes).cpu() # (B, 1, T_24)

torchaudio.save("reconstructed.wav", recon[0, :, :], 24_000)
```

## Training Details

The model was trained using the same data as the full model, with an additional distillation loss (MSE between distilled and original encoder ouputs).