{"data":{"slug":"yeyupiaoling-ppasr","name":"PPASR","tagline":"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","github_url":"https://github.com/yeyupiaoling/PPASR","owner":"yeyupiaoling","repo":"PPASR","owner_avatar_url":"https://avatars.githubusercontent.com/u/26297768?v=4","primary_language":"Python","stars":873,"forks":129,"topics":["asr","chinese","conformer","deep-learning","deepspeech2","paddlepaddle","speech","speech-recognition","speech-to-text","squeezeformer","streaming-asr"],"archived":false,"github_pushed_at":"2025-12-17T13:16:13+00:00","maintenance_label":"Slowing","url":"https://www.graphcanon.com/tools/yeyupiaoling-ppasr","markdown_url":"https://www.graphcanon.com/tools/yeyupiaoling-ppasr.md","api_url":"https://www.graphcanon.com/api/graphcanon/tools/yeyupiaoling-ppasr","graph_url":"https://www.graphcanon.com/api/graphcanon/graph?tool=yeyupiaoling-ppasr","description":"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","homepage_url":null,"license":"Apache-2.0","open_issues":1,"watchers":11,"ai_summary":null,"readme_excerpt":"# PPASR流式与非流式语音识别项目\n\nPPASR是一款基于PaddlePaddle实现的自动语音识别框架，PPASR中文名称PaddlePaddle中文语音识别（PaddlePaddle Automatic Speech Recognition），当前为V3版本，与V2版本不兼容，如果想使用V2版本，请在这个分支[V2](https://github.com/yeyupiaoling/PPASR/tree/release/2.4.x)。PPASR致力于简单，实用的语音识别项目。可部署在服务器，Nvidia Jetson设备，未来还计划支持Android等移动设备。**别忘了star**\n\n**欢迎大家扫码入知识星球或者QQ群讨论，知识星球里面提供项目的模型文件和博主其他相关项目的模型文件，也包括其他一些资源。**\n\n<div align=\"center\">\n  <img src=\"https://yeyupiaoling.cn/zsxq.jpg\" alt=\"知识星球\" width=\"400\">\n  <img src=\"https://yeyupiaoling.cn/qq.jpg\" alt=\"QQ群\" width=\"400\">\n</div>\n\n<br/>\n\n**本项目使用的环境：**\n - Anaconda 3\n - Python 3.11\n - PaddlePaddle 2.6.1\n - Windows 11 or Ubuntu 22.04\n\n\n# 在线试用\n\n**网页版：** [在线试用地址](https://tools.yeyupiaoling.cn/speech/masr)\n\n<div align=\"center\">\n  <img src=\"https://tools.yeyupiaoling.cn/static/wechat-qr/masr.jpg\" alt=\"微信小程序\" width=\"200\"><br/>\n  微信小程序\n</div>\n\n\n## 项目特点\n\n1. 支持多个语音识别模型，包含`deepspeech2`、`conformer`、`squeezeformer`、`efficient_conformer`等，每个模型都支持流式识别和非流式识别，在配置文件中`streaming`参数设置。\n2. 支持多种解码器，包含`ctc_greedy_search`、`ctc_prefix_beam_search`、`attention_rescoring`、`ctc_beam_search`等。\n3. 支持多种预处理方法，包含`fbank`、`mfcc`等。\n4. 支持多种数据增强方法，包含噪声增强、混响增强、语速增强、音量增强、重采样增强、位移增强、SpecAugmentor、SpecSubAugmentor等。\n5. 支持多种推理方法，包含短音频推理、长音频推理、流式推理、说话人分离推理等。\n6. 更多特点等待你发现。\n\n## 与V2版本的区别\n\n1. 项目结构的优化，大幅度降低的使用难度。\n2. 更换预处理的库，改用kaldi_native_fbank，在提高数据预处理的速度，同时也支持多平台。\n3. 修改token的方法，使用sentencepiece制作token，这个框架极大的降低了多种语言的处理难度，同时还使中英文混合训练成为可能。\n\n## 更新记录\n\n - 2025.03: 正式发布最终级的V3版本。\n\n\n## 视频讲解\n\n - [知识点讲解（哔哩哔哩）](https://www.bilibili.com/video/BV1Rr4y1D7iZ)\n - [流式识别的使用讲解（哔哩哔哩）](https://www.bilibili.com/video/BV1Te4y1h7KK)\n\n\n## 模型下载\n\n1. [WenetSpeech](./docs/wenetspeech.md) (10000小时，普通话) 的预训练模型列表，错误率类型为字错率（CER）：\n\n|    使用模型     | 是否为流式 | 预处理方式 |          解码方式          | test_net | test_meeting | aishell_test |   下载地址   |\n|:-----------:|:-----:|:-----:|:----------------------:|:--------:|:------------:|:------------:|:--------:|\n|  Conformer  | True  | fbank |   ctc_greedy_search    | 0.14758  |   0.19562    |   0.06925    | 加入知识星球获取 |\n|  Conformer  | True  | fbank | ctc_prefix_beam_search | 0.14689  |   0.19323    |   0.06930    | 加入知识星球获取 |\n|  Conformer  | True  | fbank |  attention_rescoring   | 0.13786  |   0.18922    |   0.06028    | 加入知识星球获取 |\n|  Conformer  | True  | fbank |    ctc_beam_search     | 0.20660  |   0.29835    |   0.05336    | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank |   ctc_greedy_search    |          |              |              | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank | ctc_prefix_beam_search |          |              |              | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank |    ctc_beam_search     |          |              |              | 加入知识星球获取 |\n\n2. [AIShell](https://openslr.magicdatatech.com/resources/33) (179小时，普通话) 的预训练模型列表，错误率类型为字错率（CER）：\n\n|    使用模型     | 是否为流式 | 预处理方式 |          解码方式          | 自带的测试集  |   下载地址   |\n|:-----------:|:-----:|:-----:|:----------------------:|:-------:|:--------:|\n|  Conformer  | True  | fbank |   ctc_greedy_search    | 0.06110 | 加入知识星球获取 |\n|  Conformer  | True  | fbank | ctc_prefix_beam_search | 0.06114 | 加入知识星球获取 |\n|  Conformer  | True  | fbank |  attention_rescoring   | 0.05412 | 加入知识星球获取 |\n|  Conformer  | True  | fbank |    ctc_beam_search     | 0.04468 | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank |   ctc_greedy_search    | 0.14134 | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank | ctc_prefix_beam_search | 0.14132 | 加入知识星球获取 |\n| DeepSpeech2 | True  | fbank |    ctc_beam_search     | 0.10598 | 加入知识星球获取 |\n\n\n3. [Librispeech](https://openslr.magicdatatech.com/resources/12) (960小时，英语) 的预训练模型列表，错误率类型为词错率（WER）：\n\n|    使用模型     | 是否为流式 | 预处理方式 |          解码方式          | 自带的测试集  |   下载地址   |\n|:-----------:|:-----:|:-----:|:----------------------:|:-------:|:--------:|\n|  Conformer  | True  | fbank |   ctc_greedy_search    | 0.07562 | 加入知识星球获取 |\n|  Conformer  | True  | fbank | ctc_prefix_beam_search | 0.07518 | 加入知识星球获取 |\n|  Conformer  | True  | fbank |  attention_rescoring   | 0.06669 | 加入知识星球获取 |\n|  Conformer  | True  | fbank |    ctc_beam_search","github_created_at":"2021-02-26T08:35:47+00:00","created_at":"2026-07-11T12:21:46.640077+00:00","updated_at":"2026-07-11T12:21:54.701451+00:00","categories":[{"slug":"speech-audio","name":"Speech & Audio","url":"https://www.graphcanon.com/categories/speech-audio","markdown_url":"https://www.graphcanon.com/categories/speech-audio.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/speech-audio"}],"tags":[{"slug":"deep-learning","name":"deep-learning"},{"slug":"asr","name":"asr"},{"slug":"chinese","name":"chinese"},{"slug":"speech","name":"speech"},{"slug":"conformer","name":"conformer"},{"slug":"paddlepaddle","name":"paddlepaddle"},{"slug":"speech-recognition","name":"speech-recognition"},{"slug":"deepspeech2","name":"deepspeech2"}],"trust":{"provenance":{"is_fork":false,"github_id":342512505,"owner_type":"User","methodology":"github_public_v1","parent_repo":null,"near_duplicate_slugs":[]},"computed_at":"2026-07-11T12:21:48.984Z","maintenance":{"label":"Slowing","score":36,"methodology":"github_public_v1","releases_90d":0,"days_since_push":205,"last_release_at":null},"security_summary":{"status":"findings","scanner":"osv@v1","low_count":6,"high_count":0,"last_scan_at":"2026-07-11T12:21:51.194Z","medium_count":0,"scan_profile":"deps","critical_count":0}},"capability_facts":{"scan":{"source":"repo_scan","observed_at":"2026-07-11T12:21:50.736Z"},"languages":{"value":["python"],"source":"github.language","observed_at":"2026-07-11T12:21:50.736Z"},"license_spdx":{"value":"Apache-2.0","source":"github.license","observed_at":"2026-07-11T12:21:50.736Z"}}}}