Metadata-Version: 2.1
Name: wvmos
Version: 1.0
Summary: This package is written for MOS score prediction based on fine-tuned wav2vec2.0 model
Home-page: https://github.com/AndreevP/wvmos
Author: Pavel Andreev
Author-email: andreev.pk@phystech.edu
Requires-Python: >=3.9.0
Description-Content-Type: text/markdown
Requires-Dist: numpy
Requires-Dist: scipy
Requires-Dist: tqdm
Requires-Dist: librosa
Requires-Dist: torch>=1.7.0
Requires-Dist: transformers>=4.12.5
Requires-Dist: pytorch-lightning>=1.5.5

# WV-MOS
MOS score prediction by fine-tuned wav2vec2.0 model

**Keywords:** MOS-Net, MB-Net, PESQ, STOI, speech quality

## Getting started
The package installation was tested with python3.9

```bash
pip install git+https://github.com/AndreevP/wvmos
```
## Inference

```python
from wvmos import get_wvmos
model = get_wvmos(cuda=True)

mos = model.calculate_one("path/to/wav/file") # infer MOS score for one audio 

mos = model.calculate_dir("path/to/dir/with/wav/files", mean=True) # infer average MOS score across .wav files in directory
```

## Citation and Acknowledgment
This work was done for the deep learning course in Skolteh university by Pavel Andreev, Nikolay Patakin, Oleg Desheulin, Alexander Kagan and Arthur Bulanbaev.
More details are described in paper https://arxiv.org/abs/2203.13086

