PPASR and PaddlePaddle-DeepSpeech
These are ecosystem siblings—PPASR is a modernized successor that consolidates the speech recognition functionality of the older PaddlePaddle-DeepSpeech project into a simplified framework supporting newer model architectures (Conformer, Squeezeformer) while maintaining the same PaddlePaddle-based foundation.
About PPASR
yeyupiaoling/PPASR
基于PaddlePaddle实现端到端中文语音识别，从入门到实战，超简单的入门案例，超实用的企业项目。支持当前最流行的DeepSpeech2、Conformer、Squeezeformer模型
Implements both streaming and non-streaming inference with multiple decoding strategies (CTC greedy search, beam search, attention rescoring) and audio preprocessing via kaldi_native_fbank for cross-platform compatibility. Uses SentencePiece tokenization to enable multilingual support including Chinese-English mixed training. Deployable across servers, Nvidia Jetson devices, and provides web/GUI interfaces alongside pre-trained models on WenetSpeech, AIShell, Librispeech, and language-specific datasets.
About PaddlePaddle-DeepSpeech
yeyupiaoling/PaddlePaddle-DeepSpeech
基于PaddlePaddle实现的语音识别，中文语音识别。项目完善，识别效果好。支持Windows，Linux下训练和预测，支持Nvidia Jetson开发板预测。
Implements the DeepSpeech2 end-to-end architecture with support for multiple decoding strategies (CTC greedy and beam search) and data augmentation techniques for various acoustic conditions. Provides pre-trained models on public datasets (AIShell, Librispeech, WenetSpeech) with character error rates as low as 5.94% on Mandarin, plus tools for long-form audio segmentation using WebRTC VAD and optional language model integration for improved accuracy.
Scores updated daily from GitHub, PyPI, and npm data. How scores work