Audio
Mozilla TTS
Mozilla's deep-learning text-to-speech library with Tacotron, Glow-TTS and neural vocoders.
★ 10.2k starsMPL-2.0
View guide →WaveRNN vocoder plus Tacotron TTS in PyTorch: train your own neural speech synthesis on LJSpeech.
WaveRNN is an open-source audio project by fatchord (fatchord/WaveRNN on GitHub). It has 2,191 GitHub stars and 685 forks, is released under the MIT license, and was last updated on GitHub on 2 July 2022.
Decide whether you want instant TTS from the pretrained models or to train on your own voice dataset.
Clone the repository and install the requirements with pip, with PyTorch and CUDA available.
For training, download the LJSpeech dataset, point the hyperparameters at your wav path, and run preprocessing.
Generate speech with the quick start script, or train Tacotron and then WaveRNN following the documented order.
Combines a Tacotron acoustic model with the WaveRNN vocoder so text goes in and waveform audio comes out.
A WaveRNN model (mixture-of-logistics output) and a Tacotron model, both trained on LJSpeech, live in the pretrained folder.
quick_start.py synthesises the default sentences, accepts custom input text, and can run unbatched for better audio quality, saving wavs and attention plots.
Preprocess, train Tacotron, optionally force a GTA dataset mid-training, train WaveRNN with or without GTA, then generate.
Dataset paths and training settings are centralised in hparams.py for easy adaptation to other corpora.
A samples page linked from the README lets you hear output quality before installing anything.
Make sure PyTorch 1 with CUDA is installed first, per the README.
git clone https://github.com/fatchord/WaveRNN.git
cd WaveRNN
pip install -r requirements.txtGenerate the default sentences into a quick_start output folder.
python quick_start.pyPass your own text, optionally unbatched for higher quality.
python quick_start.py -u --input_text "What will happen if I run this command?"It is maintained by fatchord, building on the WaveRNN and Tacotron papers and earlier open implementations credited in the README.
Yes, it is MIT-licensed, including use of the bundled pretrained models.
No. Everything runs locally in PyTorch; there are no hosted APIs or keys.
HiFi-GAN is a GAN vocoder focused on mel inversion; WaveRNN pairs its autoregressive vocoder with a full Tacotron TTS front end and its own training pipeline in one repo.
Clone the repo, pip install -r requirements.txt, and run python quick_start.py to hear the pretrained system immediately.
Mozilla's deep-learning text-to-speech library with Tacotron, Glow-TTS and neural vocoders.
Unofficial PyTorch implementation of VALL-E, the neural codec language model for zero-shot TTS
A PyTorch-based Speech Toolkit
Offline text-to-speech for Python — uses your operating system's own voices, no internet needed
Turn books and documents into natural speech with the Kokoro TTS model.
Deep learning toolkit for text-to-speech, battle-tested in research and production
The most-starred other Audio projects in the LibreAI catalogue — the realistic shortlist if WaveRNN is not the right fit.
利用 AI 大模型和自动化工作流，根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.
Robust Speech Recognition via Large-Scale Weak Supervision
1 min voice data can also be used to train a good TTS model! (few shot voice cloning)
Clone a voice from 5 seconds of audio and generate arbitrary speech in real time
Port of OpenAI's Whisper model in C/C++
Free and open — grab the source, star the repo, and start creating.