PaddleSpeech is Baidu’s open-source speech toolkit covering the full voice stack: automatic speech recognition, text-to-speech, voice cloning, speaker verification, speech translation and more. It won the NAACL 2022 Best Demo Award and has collected about 12,700 stars on GitHub.
Most speech projects give you one model; PaddleSpeech is an end-to-end toolkit for building voice products. Its ASR systems include streaming Conformer models and self-supervised approaches such as wav2vec 2.0, with punctuation restoration for readable transcripts, while its TTS pipeline combines an acoustic model with a text frontend and vocoders for natural streaming synthesis, including voice cloning from short samples. Around those cores sit the supporting capabilities real products need: keyword spotting for wake words, speaker verification, end-to-end speech translation and sound classification. The toolkit is particularly strong on Mandarin as well as English, reflecting its Baidu origins, and it ships with deployment paths from research notebooks to serving. For teams in the Paddle ecosystem, it plays the same role for speech that PaddleNLP plays for text.
PaddleSpeech is free and open source under the Apache-2.0 license. You run the models on your own hardware with no per-minute or per-character fees.
PaddleSpeech is for speech engineers and product teams building transcription, voice assistants, dubbing or call-center analytics — especially where strong Chinese speech support matters — who want a complete, self-hosted alternative to commercial speech APIs.