https://github.com/laion-ai/text-to-speech
https://github.com/laion-ai/text-to-speech
Last synced: about 1 month ago
JSON representation
- Host: GitHub
- URL: https://github.com/laion-ai/text-to-speech
- Owner: LAION-AI
- License: apache-2.0
- Created: 2023-10-18T06:09:40.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-11-04T10:10:21.000Z (over 1 year ago)
- Last Synced: 2025-05-07T18:13:31.147Z (about 1 month ago)
- Language: Python
- Size: 60.5 KB
- Stars: 59
- Watchers: 10
- Forks: 6
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
## Audio pipeline for TTS Datasets
This project aims to provide high level APIs for various feature engineering techniques for processing audio files. The implementation follows a modularized and config based approach, so any dataset processing pipeline can be built and managed using the same.
### Creating new pipeline
Creat an yaml file under `config/pipelines/` directory with the following structure
```
pipeline:
loader:
target: manager.Downloader
args:
configs:
- config/datasets/data-config.yaml
save_dir: raw_data/yt_data
manager:
target: manager.YoutubeRunner
processors:
- name: chunking
target: modules.AudioChunking
args:
model_choice: pydub_chunking
- name: denoise_audio
target: modules.DenoiseAudio
args:
model_choice: meta_denoiser_dns48
- name: audio_superres
target: modules.SuperResAudio
args:
model_choice: voicefixer
...
```
**Pipeline Schema**
- **Loader**: The entry point for fetching data from various sources like S3, local systems, or blob storage.
- **Manager**: Specifies the manager class responsible for running the pipeline.
- **Processors**: An ordered list of processors to apply for feature extraction or other manipulations.
If new feature extractors or manager are required for your needs, check the `modules/` directory for understanding the structure and create or update the objects as needed.
### Run pipleines
```
python workers/pipeline.py --configs
```
## Acknowledgements
credit a few of the amazing folks in the community that have helped to this happen:
- [bud-studio](https://bud.studio/) - For providing a initial framework