Repolode
Submit
PaddlePaddle

PaddlePaddle/PaddleSpeech

An open-source speech toolkit built on PaddlePaddle for speech recognition, text-to-speech, speaker verification, keyword spotting and speech translation.

View on GitHubWebsite
Stars
12,704
Forks
1,952
Open issues
277
Last commit
3 months ago

What is it?

PaddleSpeech is a speech and audio toolkit on the PaddlePaddle deep learning framework. It provides ready-to-use command-line tools and Python APIs for automatic speech recognition, text-to-speech, audio classification, speaker verification, speaker diarization, keyword spotting, punctuation restoration and English-to-Chinese speech translation.

The toolkit ships with pretrained models and training recipes for many architectures, including DeepSpeech2, Conformer and Transformer-based ASR, FastSpeech2, VITS and HiFiGAN for TTS, and ECAPA-TDNN for speaker tasks. It includes a Chinese text frontend with text normalization and grapheme-to-phoneme conversion, and offers both offline and streaming ASR and TTS servers. Models can be exported to ONNX for TTS.

The project covers the full pipeline from data preparation and training to inference and deployment, and includes C++ runtime examples and Android and ARM Linux demos.

Key features

  • 01Speech recognition with streaming and non-streaming models, plus punctuation restoration.
  • 02Text-to-speech with a Chinese text frontend, multiple acoustic models and vocoders.
  • 03Speaker verification, speaker diarization and audio classification.
  • 04Keyword spotting and English-to-Chinese speech translation.
  • 05Command-line tools, Python API, and HTTP/streaming servers for deployment.

Use cases

  • Adding speech-to-text or subtitle generation to Chinese-language applications.
  • Building text-to-speech voices for virtual hosts, audiobooks or voice assistants.
  • Running speaker verification or voiceprint recognition in a backend service.
  • Researching or fine-tuning speech models on datasets such as AISHELL and LibriSpeech.

Things to know

  • Built on the PaddlePaddle framework, so that framework must be installed to use it.
  • The README excerpt provided does not include installation commands.

In collections

AI video & audio tools

Topics

#speech-recognition#text-to-speech#speaker-verification#asr#tts#audio#deep-learning#paddlepaddle#code-switch#conformer

FAQ

What is PaddleSpeech?
An open-source speech toolkit built on PaddlePaddle for speech recognition, text-to-speech, speaker verification, keyword spotting and speech translation. PaddleSpeech is a speech and audio toolkit on the PaddlePaddle deep learning framework. It provides ready-to-use command-line tools and Python APIs for automatic speech recognition, text-to-speech, audio classification, speaker verification, speaker diarization, keyword spotting, punctuation restoration and English-to-Chinese speech translation.
Is PaddleSpeech free to use? What license does it have?
PaddleSpeech is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is PaddleSpeech still maintained?
Updated within 6 months. The last commit to PaddleSpeech was on Aug 12, 2026.
What can I use PaddleSpeech for?
Adding speech-to-text or subtitle generation to Chinese-language applications. Building text-to-speech voices for virtual hosts, audiobooks or voice assistants. Running speaker verification or voiceprint recognition in a backend service. Researching or fine-tuning speech models on datasets such as AISHELL and LibriSpeech.
Where can I get PaddleSpeech?
The source code and installation instructions are on GitHub: https://github.com/PaddlePaddle/PaddleSpeech

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects