An unofficial PyTorch implementation of VALL-E
Go to file
2023-08-20 13:39:58 -05:00
data nightmare of spaghetti that might break compat; mechanism to increase RVQ bins of an existing model without retraining, keeps sampled proms/resps at max RVQ level and trim off excess levels according to what model receives them, some other things I already forgot (I really hope no one else has weights being baked right now) 2023-08-19 15:06:33 -05:00
scripts Rewrite init 2023-08-02 21:53:35 +00:00
vall_e added ability to export weights mid-training to avoid CBT to yank the weights while the training script is running 2023-08-20 13:39:58 -05:00
.gitignore set model to eval when inferencing (very important) 2023-08-05 04:29:05 +00:00
LICENSE Rewrite init 2023-08-02 21:53:35 +00:00
README.md validated that SpeechX tasks cse and nse works, added a method to test each task by invoking python3 -m vall_e.data --action=tasks --tasks='sr,se,cse,nse' 2023-08-19 09:50:07 -05:00
setup.py Fixed an issue with having fairseq installed at all will brick logging 2023-08-02 22:57:10 -05:00
vall-e.png Rewrite init 2023-08-02 21:53:35 +00:00

VALL'E

An unofficial PyTorch implementation of VALL-E, based on the EnCodec tokenizer.

Note this is highly experimental. While I've seem to have audited and tighened down as much as I can, I'm still trying to produce a decent model out of it. You're free to train your own model if you happen to have the massive compute for it, but it's quite the beast to properly feed. This README won't get much love until I truly nail out a quasi-decent model.

Note You can follow along with my pseudo-blog in an issue here. I currently have a dataset clocking in at 3400+ trimmed hours.

Requirements

If your config YAML has the training backend set to deepspeed, you will need to have a GPU that DeepSpeed has developed and tested against, as well as a CUDA or ROCm compiler pre-installed to install this package.

Install

pip install git+https://git.ecker.tech/mrq/vall-e

Or you may clone by:

git clone --recurse-submodules https://git.ecker.tech/mrq/vall-e.git

I've tested this repo under Python versions 3.10.9 and 3.11.3.

Try Me

To quickly try it out, you can choose between the following modes:

  • AR only: python -m vall_e.models.ar yaml="./data/config.yaml"
  • NAR only: python -m vall_e.models.nar yaml="./data/config.yaml"
  • AR+NAR: python -m vall_e.models.base yaml="./data/config.yaml"

Each model file has a barebones trainer and inference routine.

Train

Training is very dependent on:

  • the quality of your dataset.
  • how much data you have.
  • the bandwidth you quantized your audio to.

Notices

Modifying prom_levels or tasks For a Model

If you're wanting to increase the prom_levels for a given model, or increase the tasks levels a model accepts, you will need to export your weights and set train.load_state_dict to True in your configuration YAML.

Leverage Your Own Dataset

Note It is highly recommended to utilize mrq/ai-voice-cloning with --tts-backend="vall-e" to handle transcription and dataset preparations.

  1. Put your data into a folder, e.g. ./data/custom. Audio files should be named with the suffix .wav and text files with .txt.

  2. Quantize the data: python -m vall_e.emb.qnt ./data/custom

  3. Generate phonemes based on the text: python -m vall_e.emb.g2p ./data/custom

  4. Customize your configuration and define the dataset by modifying ./data/config.yaml. Refer to ./vall_e/config.py for details. If you want to choose between different model presets, check ./vall_e/models/__init__.py.

If you're interested in creating an HDF5 copy of your dataset, simply invoke: python -m vall_e.data --action='hdf5' yaml='./data/config.yaml'

  1. Train the AR and NAR models using the following scripts: python -m vall_e.train yaml=./data/config.yaml

You may quit your training any time by just typing quit in your CLI. The latest checkpoint will be automatically saved.

Dataset Formats

Two dataset formats are supported:

  • the standard way:
    • data is stored under ${speaker}/${id}.phn.txt and ${speaker}/${id}.qnt.pt
  • using an HDF5 dataset:
    • you can convert from the standard way with the following command: python3 -m vall_e.data yaml="./path/to/your/config.yaml"
    • this will shove everything into a single HDF5 file and store some metadata alongside (for now, the symbol map generated, and text/audio lengths)
    • be sure to also define use_hdf5 in your config YAML.

Export

Both trained models can be exported, but is only required if loading them on systems without DeepSpeed for inferencing (Windows systems). To export the models, run: python -m vall_e.export yaml=./data/config.yaml.

This will export the latest checkpoints under ./data/ckpt/ar-retnet-2/fp32.pth and ./data/ckpt/nar-retnet-2/fp32.pth to be loaded on any system with PyTorch.

Synthesis

To synthesize speech, invoke either (if exported the models): python -m vall_e <text> <ref_path> <out_path> --ar-ckpt ./models/ar.pt --nar-ckpt ./models/nar.pt or python -m vall_e <text> <ref_path> <out_path> yaml=<yaml_path>

Some additional flags you can pass are:

  • --max-ar-steps: maximum steps for inferencing through the AR model. Each second is 75 steps.
  • --ar-temp: sampling temperature to use for the AR pass. During experimentation, 0.95 provides the most consistent output.
  • --nar-temp: sampling temperature to use for the NAR pass. During experimentation, 0.2 provides the most clean output.
  • --device: device to use (default: cuda, examples: cuda:0, cuda:1, cpu)

To-Do

  • reduce load time for creating / preparing dataloaders.
  • properly pass in modules names to weight_quantization and activation_quantization.
  • train and release a model.
  • extend to multiple languages (VALL-E X) and extend to SpeechX features.

Notice

  • EnCodec is licensed under CC-BY-NC 4.0. If you use the code to generate audio quantization or perform decoding, it is important to adhere to the terms of their license.

Unless otherwise credited/noted, this repository is licensed under AGPLv3.

Citations

@article{wang2023neural,
  title={Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers},
  author={Wang, Chengyi and Chen, Sanyuan and Wu, Yu and Zhang, Ziqiang and Zhou, Long and Liu, Shujie and Chen, Zhuo and Liu, Yanqing and Wang, Huaming and Li, Jinyu and others},
  journal={arXiv preprint arXiv:2301.02111},
  year={2023}
}
@article{defossez2022highfi,
  title={High Fidelity Neural Audio Compression},
  author={Défossez, Alexandre and Copet, Jade and Synnaeve, Gabriel and Adi, Yossi},
  journal={arXiv preprint arXiv:2210.13438},
  year={2022}
}