Metadata-Version: 2.4
Name: frechet_audio_distance
Version: 0.3.3
Summary: A lightweight library of Frechet Audio Distance calculation.
Author-email: Hao Hao Tan <helloharry66@gmail.com>
License-Expression: MIT
Project-URL: Homepage, https://github.com/gudgud96/frechet-audio-distance
Classifier: Programming Language :: Python :: 3
Classifier: Operating System :: OS Independent
Requires-Python: >=3.10
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: numpy>=2.0.0; python_version >= "3.13"
Requires-Dist: numpy<2.0.0; python_version < "3.13"
Requires-Dist: torch
Requires-Dist: scipy
Requires-Dist: tqdm
Requires-Dist: soundfile
Requires-Dist: resampy
Requires-Dist: torchlibrosa
Requires-Dist: torchvision
Requires-Dist: laion_clap
Requires-Dist: transformers
Requires-Dist: torchaudio
Requires-Dist: encodec
Dynamic: license-file

## Frechet Audio Distance in PyTorch

A lightweight library of Frechet Audio Distance (FAD) calculation.

Currently, we support:
- FAD score, with embeddings from:
    - `VGGish` by [S. Hershey et al.](https://arxiv.org/abs/1812.08466)
    
    - `PANN` by [Kong et al.](https://arxiv.org/abs/1912.10211)
    - `CLAP` by [Wu et al.](https://arxiv.org/abs/2211.06687)
    - `EnCodec` by [Defossez et al.](https://arxiv.org/pdf/2210.13438.pdf)

- CLAP score, for text and audio matching

### Installation

`pip install frechet_audio_distance`

### Example

#### For FAD:

```python
from frechet_audio_distance import FrechetAudioDistance

# to use `vggish`
frechet = FrechetAudioDistance(
    model_name="vggish",
    sample_rate=16000,
    use_pca=False, 
    use_activation=False,
    verbose=False
)
# to use `PANN`
frechet = FrechetAudioDistance(
    model_name="pann",
    sample_rate=16000,
    use_pca=False, 
    use_activation=False,
    verbose=False
)
# to use `CLAP`
frechet = FrechetAudioDistance(
    model_name="clap",
    sample_rate=48000,
    submodel_name="630k-audioset",  # for CLAP only
    verbose=False,
    enable_fusion=False,            # for CLAP only
)
# to use `EnCodec`
frechet = FrechetAudioDistance(
    model_name="encodec",
    sample_rate=48000,
    channels=2,
    verbose=False,
)

fad_score = frechet.score(
    "/path/to/background/set", 
    "/path/to/eval/set", 
    dtype="float32"
)
```

You can also have a look at [this notebook](https://github.com/gudgud96/frechet-audio-distance/blob/main/test/test_all.ipynb) for a better understanding of how each model is used.

#### For CLAP score:

```python
from frechet_audio_distance import CLAPScore

clap = CLAPScore(
    submodel_name="630k-audioset",
    verbose=True,
    enable_fusion=False,
)

clap_score = clap.score(
    text_path="./text1/text.csv",
    audio_dir="./audio1",
    text_column="caption",
)
```

For more info, kindly refer to [this notebook](https://github.com/gudgud96/frechet-audio-distance/blob/main/test/test_clap_score.ipynb).

### Save pre-computed embeddings

When computing the Frechet Audio Distance, you can choose to save the embeddings for future use. 

This capability not only ensures consistency across evaluations but can also significantly reduce computation time, especially if you're evaluating multiple times using the same dataset.

```python
# Specify the paths to your saved embeddings
background_embds_path = "/path/to/saved/background/embeddings.npy"
eval_embds_path = "/path/to/saved/eval/embeddings.npy"

# Compute FAD score while reusing the saved embeddings (or saving new ones if paths are provided and embeddings don't exist yet)
fad_score = frechet.score(
    "/path/to/background/set",
    "/path/to/eval/set",
    background_embds_path=background_embds_path,
    eval_embds_path=eval_embds_path,
    dtype="float32"
)
```

### Result validation

**Test 1: Distorted sine waves on vggish** (as provided [here](https://github.com/google-research/google-research/blob/master/frechet_audio_distance/gen_test_files.py#L86)) [[notes](https://jexrj22lgy.larksuite.com/docx/Vat2dr8Aqonim6xmE6nuoBVZsUe)]

FAD scores comparison w.r.t. to original implementation in `google-research/frechet-audio-distance`

|                              |   baseline vs test1   |     baseline vs test2    |
|:----------------------------:|:---------------------:|:------------------------:|
|        `google-research`     |          12.4375      |           4.7680         |
|    `frechet_audio_distance`  |          12.7398      |           4.9815         |

**Test 2: Distorted sine waves on PANN**

|                              |   baseline vs test1   |     baseline vs test2    |
|:----------------------------:|:---------------------:|:------------------------:|
|    `frechet_audio_distance`  |        0.000465       |          0.00008594      |

### To contribute

Contributions are welcomed! Kindly raise a PR and ensure that all CI checks are passed.

NOTE: For now, the CI only checks for `vggish` as `PANN` takes a long time to download.

### References

VGGish in PyTorch: https://github.com/harritaylor/torchvggish

Frechet distance implementation: https://github.com/mseitzer/pytorch-fid

Frechet Audio Distance paper: https://arxiv.org/abs/1812.08466

PANN paper: https://arxiv.org/abs/1912.10211
