thunnai/audiobox-aesthetics

audiobox-aesthetics

Unified automatic quality assessment for speech, music, and sound.

Read our paper here.

Installation

1. 1.Install via pip

    pip install audiobox_aesthetics

1. 1.Install directly from source

This repository requires Python 3.9 and Pytorch 2.2 or greater. To install, you can clone this repo and run:

    pip install -e .

Pre-trained Models

Usage

How to run prediction:

1. 1.Create a jsonl files with the following format

    {"path":"/path/to/a.wav"}
    {"path":"/path/to/b.wav"}
    ...
    {"path":"/path/to/z.wav"}

or if you only want to predict aesthetic score from certain timestamp

    {"path":"/path/to/a.wav", "start_time":0, "end_time": 5}
    {"path":"/path/to/b.wav", "start_time":3, "end_time": 10}

and save it as input.jsonl

1. 1.Run following command

    audio-aes input.jsonl --ckpt "/path/to/checkpoint.pt" > output.jsonl

If path for ckpt didn't exist or you haven't download the checkpoint, the script will try to download it automatically.

1. 1.Output file will contains same number of rows as input.jsonl. Each rows contains 4 axes prediction with JSON-formatted dictionary. Check following table for more info: Axes name | Full name |---|---| CE | Content Enjoyment CU | Content Usefulness PC | Production Complexity PQ | Production Quality

Output line example:

    {"CE": 5.146, "CU": 5.779, "PC": 2.148, "PQ": 7.220}

1. 1.(Extra) If you want to extract only one axis (i.e. CE), post-process the output file with following command using jq utility:

## Evaluation dataset
We released our evaluation dataset consisted of 4 axes of aesthetic annotation scores. 

Here, we show an example on how to read and re-map each annotation to the actual audio file.

{ "datapath": "/yourpath/LibriTTS/train-clean-100/1363/139304/1363139304000011000000.wav", "ProductionQuality": [8.0, 8.0, 8.0, 8.0, 8.0, 9.0, 8.0, 5.0, 8.0, 8.0], "ProductionComplexity": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "ContentEnjoyment": [8.0, 6.0, 8.0, 5.0, 8.0, 8.0, 8.0, 6.0, 8.0, 6.0], "Content_Usefulness": [8.0, 6.0, 8.0, 7.0, 8.0, 9.0, 8.0, 6.0, 10.0, 7.0] }

1. Recognize the dataset name from data_path. In the example, it is LibriTTS.
2. Replace "/your_path/" into your downloaded LibriTTS directory. 
3. Each axes contains 10 scores annotated by 10 different human annotators.

data_path | URL
|---|---|
LibriTTS |  https://openslr.org/60/
cv-corpus-13.0-2023-03-09 | https://commonvoice.mozilla.org/en/datasets
EARS | https://sp-uhh.github.io/ears_dataset/
MUSDB18 | https://sigsep.github.io/datasets/musdb.html
musiccaps | https://www.kaggle.com/datasets/googleai/musiccaps
(audioset) unbalanced_train_segments | https://research.google.com/audioset/dataset/index.html 
PAM | https://zenodo.org/records/10737388

## License
The majority of audiobox-aesthetics is licensed under CC-BY 4.0, as found in the LICENSE file.
However, portions of the project are available under separate license terms: [https://github.com/microsoft/unilm](https://github.com/microsoft/unilm) is licensed under MIT license.

## Citation
If you found this repository useful, please use the following BibTeX entry. (will include arXiv link soon)

@article{tjandra2025aes, title={Meta Audiobox Aesthetics: Unified Automatic Quality Assessment for Speech, Music, and Sound}, author={Tjandra, Andros and Wu, Yi-Chiao and Guo, Baishan and Hoffman, John and Ellis, Brian and Vyas, Apoorv and Shi, Bowen and Chen, Sanyuan and Le, Matt and Zacharov, Nick and Wood, Carleigh and Lee, Ann and Hsu, Wei-ning}, publisher={Meta AI}, year={2025}, url={https://ai.meta.com/research/publications/meta-audiobox-aesthetics-unified-automatic-quality-assessment-for-speech-music-and-sound/} }

## Acknowledgements

Part of model code are copied from [https://github.com/microsoft/unilm/tree/master/wavlm](WavLM).