x-lance / f5-tts

F5-TTS, the new state-of-the-art in open source voice cloning

  • Public
  • 18.6K runs
  • GitHub
  • Weights
  • Paper
  • License
Run with an API
Playground API Examples README Versions

Run time and cost

This model costs approximately $0.10 to run on Replicate, or 10 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia A100 (80GB) GPU hardware. Predictions typically complete within 72 seconds. The predict time for this model varies significantly based on the inputs.

Readme

F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching

F5-TTS: Diffusion Transformer with ConvNeXt V2, faster trained and inference.

E2 TTS: Flat-UNet Transformer, closest reproduction.

Sway Sampling: Inference-time flow step sampling strategy, greatly improves performance

Installation

Clone the repository:

git clone https://github.com/SWivid/F5-TTS.git
cd F5-TTS

Install packages:

pip install -r requirements.txt

Install torch with your CUDA version, e.g. :

pip install torch==2.3.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118
pip install torchaudio==2.3.0+cu118 --extra-index-url https://download.pytorch.org/whl/cu118

Note: install numpy with version < 2.x, e.g. pip install numpy==1.22.0.

Prepare Dataset

Example data processing scripts for Emilia and Wenetspeech4TTS, and you may tailor your own one along with a Dataset class in model/dataset.py.

# prepare custom dataset up to your need
# download corresponding dataset first, and fill in the path in scripts

# Prepare the Emilia dataset
python scripts/prepare_emilia.py

# Prepare the Wenetspeech4TTS dataset
python scripts/prepare_wenetspeech4tts.py

Training

Once your datasets are prepared, you can start the training process.

# setup accelerate config, e.g. use multi-gpu ddp, fp16
# will be to: ~/.cache/huggingface/accelerate/default_config.yaml     
accelerate config
accelerate launch test_train.py

An initial guidance on Finetuning #57.

Inference

To run inference with pretrained models, download the checkpoints from 🤗 Hugging Face.

Currently support up to 30s generation, which is the TOTAL length of prompt audio and the generated. Batch inference with chunks is supported by Gradio APP now. - To avoid possible inference failures, make sure you have seen through the following instructions. - A longer prompt audio allows shorter generated output. The part longer than 30s cannot be generated properly. Consider split your text and do several separate inferences or leverage the local Gradio APP which enables a batch inference with chunks. - Uppercased letters will be uttered letter by letter, so use lowercased letters for normal words. - Add some spaces (blank: ” “) or punctuations (e.g. “,” “.”) to explicitly introduce some pauses. If first few words skipped in code-switched generation (cuz different speed with different languages), this might help.

Single Inference

You can test single inference using the following command. Before running the command, modify the config up to your need.

# modify the config up to your need,
# e.g. fix_duration (the total length of prompt + to_generate, currently support up to 30s)
#      nfe_step     (larger takes more time to do more precise inference ode)
#      ode_method   (switch to 'midpoint' for better compatibility with small nfe_step, )
#                   ( though 'midpoint' is 2nd-order ode solver, slower compared to 1st-order 'Euler')
python test_infer_single.py

Speech Editing

To test speech editing capabilities, use the following command.

python test_infer_single_edit.py

Gradio App

You can launch a Gradio app (web interface) to launch a GUI for inference.

First, make sure you have the dependencies installed (pip install -r requirements.txt). Then, install the Gradio app dependencies:

pip install -r requirements_gradio.txt

After installing the dependencies, launch the app (will load ckpt from Huggingface, you may set ckpt_path to local file in gradio_app.py):

python gradio_app.py

You can specify the port/host:

python gradio_app.py --port 7860 --host 0.0.0.0

Or launch a share link:

python gradio_app.py --share

Evaluation

Prepare Test Datasets

  1. Seed-TTS test set: Download from seed-tts-eval.
  2. LibriSpeech test-clean: Download from OpenSLR.
  3. Unzip the downloaded datasets and place them in the data/ directory.
  4. Update the path for the test-clean data in test_infer_batch.py
  5. Our filtered LibriSpeech-PC 4-10s subset is already under data/ in this repo

Batch Inference for Test Set

To run batch inference for evaluations, execute the following commands:

# batch inference for evaluations
accelerate config  # if not set before
bash test_infer_batch.sh

Download Evaluation Model Checkpoints

  1. Chinese ASR Model: Paraformer-zh
  2. English ASR Model: Faster-Whisper
  3. WavLM Model: Download from Google Drive.

Objective Evaluation

Some Notes

For faster-whisper with CUDA 11:

pip install --force-reinstall ctranslate2==3.24.0

(Recommended) To avoid possible ASR failures, such as abnormal repetitions in output:

pip install faster-whisper==0.10.1

Update the path with your batch-inferenced results, and carry out WER / SIM evaluations:

# Evaluation for Seed-TTS test set
python scripts/eval_seedtts_testset.py

# Evaluation for LibriSpeech-PC test-clean (cross-sentence)
python scripts/eval_librispeech_test_clean.py

Acknowledgements

Citation

@article{chen-etal-2024-f5tts,
      title={F5-TTS: A Fairytaler that Fakes Fluent and Faithful Speech with Flow Matching}, 
      author={Yushen Chen and Zhikang Niu and Ziyang Ma and Keqi Deng and Chunhui Wang and Jian Zhao and Kai Yu and Xie Chen},
      journal={arXiv preprint arXiv:2410.06885},
      year={2024},
}

License

Our code is released under MIT License.