PaddlePaddle DeepSpeech Save

Easy-to-use Speech Toolkit including Self-Supervised Learning model, SOTA/Streaming ASR with punctuation, Streaming TTS with text frontend, Speaker Verification System, End-to-End Speech Translation and Keyword Spotting. Won NAACL2022 Best Demo Award.

Project README


PaddleSpeech is an open-source toolkit on PaddlePaddle platform for a variety of critical tasks in speech and audio, with the state-of-art and influential models.

PaddleSpeech won the NAACL2022 Best Demo Award, please check out our paper on Arxiv.

Speech Recognition
Input Audio Recognition Result

I knocked at the door on the ancient side of the building.

Speech Translation (English to Chinese)
Input Audio Translations Result

我 在 这栋 建筑 的 古老 门上 敲门。
Input Text Synthetic Audio
Life was like a box of chocolates, you never know what you're gonna get.
大家好,我是 parrot 虚拟老师,我们来读一首诗,我与春风皆过客,I and the spring breeze are passing by,你携秋水揽星河,you take the autumn water to take the galaxy。

For more synthesized audios, please refer to PaddleSpeech Text-to-Speech samples.

Punctuation Restoration
Input Text Output Text
今天的天气真不错啊你下午有空吗我想约你一起去吃饭 今天的天气真不错啊!你下午有空吗?我想约你一起去吃饭。


Via the easy-to-use, efficient, flexible and scalable implementation, our vision is to empower both industrial application and academic research, including training, inference & testing modules, and deployment process. To be more specific, this toolkit features at:

  • 📦 Ease of Use: low barriers to install, CLI, Server, and Streaming Server is available to quick-start your journey.
  • 🏆 Align to the State-of-the-Art: we provide high-speed and ultra-lightweight models, and also cutting-edge technology.
  • 🏆 Streaming ASR and TTS System: we provide production ready streaming asr and streaming tts system.
  • 💯 Rule-based Chinese frontend: our frontend contains Text Normalization and Grapheme-to-Phoneme (G2P, including Polyphone and Tone Sandhi). Moreover, we use self-defined linguistic rules to adapt Chinese context.
  • 📦 Varieties of Functions that Vitalize both Industrial and Academia:
    • 🛎️ Implementation of critical audio tasks: this toolkit contains audio functions like Automatic Speech Recognition, Text-to-Speech Synthesis, Speaker Verfication, KeyWord Spotting, Audio Classification, and Speech Translation, etc.
    • 🔬 Integration of mainstream models and datasets: the toolkit implements modules that participate in the whole pipeline of the speech tasks, and uses mainstream datasets like LibriSpeech, LJSpeech, AIShell, CSMSC, etc. See also model list for more details.
    • 🧩 Cascaded models application: as an extension of the typical traditional audio tasks, we combine the workflows of the aforementioned tasks with other fields like Natural language processing (NLP) and Computer Vision (CV).

Recent Update


  • Scan the QR code below with your Wechat, you can access to official technical exchange group and get the bonus ( more than 20GB learning materials, such as papers, codes and videos ) and the live link of the lessons. Look forward to your participation.


We strongly recommend our users to install PaddleSpeech in Linux with python>=3.8 and paddlepaddle<=2.5.1. Some new versions of Paddle do not have support for adaptation in PaddleSpeech, so currently only versions 2.5.1 and earlier can be supported.

Dependency Introduction

  • gcc >= 4.8.5
  • paddlepaddle <= 2.5.1
  • python >= 3.8
  • OS support: Linux(recommend), Windows, Mac OSX

PaddleSpeech depends on paddlepaddle. For installation, please refer to the official website of paddlepaddle and choose according to your own machine. Here is an example of the cpu version.

pip install paddlepaddle -i

You can also specify the version of paddlepaddle or install the develop version.

# install 2.4.1 version. Note, 2.4.1 is just an example, please follow the minimum dependency of paddlepaddle for your selection
pip install paddlepaddle==2.4.1 -i
# install develop version
pip install paddlepaddle==0.0.0 -f

There are two quick installation methods for PaddleSpeech, one is pip installation, and the other is source code compilation (recommended).

pip install

pip install pytest-runner
pip install paddlespeech

source code compilation

git clone
cd PaddleSpeech
pip install pytest-runner
pip install .

For more installation problems, such as conda environment, librosa-dependent, gcc problems, kaldi installation, etc., you can refer to this installation document. If you encounter problems during installation, you can leave a message on #2150 and find related problems

Quick Start

Developers can have a try of our models with PaddleSpeech Command Line or Python. Change --input to test your own audio/text and support 16k wav format audio.

You can also quickly experience it in AI Studio 👉🏻 PaddleSpeech API Demo

Test audio sample download

wget -c
wget -c

Automatic Speech Recognition

 (Click to expand)Open Source Speech Recognition

command line experience

paddlespeech asr --lang zh --input zh.wav

Python API experience

>>> from paddlespeech.cli.asr.infer import ASRExecutor
>>> asr = ASRExecutor()
>>> result = asr(audio_file="zh.wav")
>>> print(result)


 Open Source Speech Synthesis

Output 24k sample rate wav format audio

command line experience

paddlespeech tts --input "你好,欢迎使用百度飞桨深度学习框架!" --output output.wav

Python API experience

>>> from paddlespeech.cli.tts.infer import TTSExecutor
>>> tts = TTSExecutor()
>>> tts(text="今天天气十分不错。", output="output.wav")

Audio Classification

 An open-domain sound classification tool

Sound classification model based on 527 categories of AudioSet dataset

command line experience

paddlespeech cls --input zh.wav

Python API experience

>>> from paddlespeech.cli.cls.infer import CLSExecutor
>>> cls = CLSExecutor()
>>> result = cls(audio_file="zh.wav")
>>> print(result)
Speech 0.9027186632156372

Voiceprint Extraction

 Industrial-grade voiceprint extraction tool

command line experience

paddlespeech vector --task spk --input zh.wav

Python API experience

>>> from paddlespeech.cli.vector import VectorExecutor
>>> vec = VectorExecutor()
>>> result = vec(audio_file="zh.wav")
>>> print(result) # 187维向量
[ -0.19083306   9.474295   -14.122263    -2.0916545    0.04848729
   4.9295826    1.4780062    0.3733844   10.695862     3.2697146
  -4.48199     -0.6617882   -9.170393   -11.1568775   -1.2358263 ...]

Punctuation Restoration

 Quick recovery of text punctuation, works with ASR models

command line experience

paddlespeech text --task punc --input 今天的天气真不错啊你下午有空吗我想约你一起去吃饭

Python API experience

>>> from paddlespeech.cli.text.infer import TextExecutor
>>> text_punc = TextExecutor()
>>> result = text_punc(text="今天的天气真不错啊你下午有空吗我想约你一起去吃饭")

Speech Translation

 End-to-end English to Chinese Speech Translation Tool

Use pre-compiled kaldi related tools, only support experience in Ubuntu system

command line experience

paddlespeech st --input en.wav

Python API experience

>>> from import STExecutor
>>> st = STExecutor()
>>> result = st(audio_file="en.wav")
['我 在 这栋 建筑 的 古老 门上 敲门 。']

Quick Start Server

Developers can have a try of our speech server with PaddleSpeech Server Command Line.

You can try it quickly in AI Studio (recommend): SpeechServer

Start server

paddlespeech_server start --config_file ./demos/speech_server/conf/application.yaml

Access Speech Recognition Services

paddlespeech_client asr --server_ip --port 8090 --input input_16k.wav

Access Text to Speech Services

paddlespeech_client tts --server_ip --port 8090 --input "您好,欢迎使用百度飞桨语音合成服务。" --output output.wav

Access Audio Classification Services

paddlespeech_client cls --server_ip --port 8090 --input input.wav

For more information about server command lines, please see: speech server demos

Quick Start Streaming Server

Developers can have a try of streaming asr and streaming tts server.

Start Streaming Speech Recognition Server

paddlespeech_server start --config_file ./demos/streaming_asr_server/conf/application.yaml

Access Streaming Speech Recognition Services

paddlespeech_client asr_online --server_ip --port 8090 --input input_16k.wav

Start Streaming Text to Speech Server

paddlespeech_server start --config_file ./demos/streaming_tts_server/conf/tts_online_application.yaml

Access Streaming Text to Speech Services

paddlespeech_client tts_online --server_ip --port 8092 --protocol http --input "您好,欢迎使用百度飞桨语音合成服务。" --output output.wav

For more information please see: streaming asr and streaming tts

Model List

PaddleSpeech supports a series of most popular models. They are summarized in released models and attached with available pretrained models.

Speech-to-Text contains Acoustic Model, Language Model, and Speech Translation, with the following details:

Speech-to-Text Module Type Dataset Model Type Example
Speech Recogination Aishell DeepSpeech2 RNN + Conv based Models deepspeech2-aishell
Transformer based Attention Models u2.transformer.conformer-aishell
Librispeech Transformer based Attention Models deepspeech2-librispeech / transformer.conformer.u2-librispeech / transformer.conformer.u2-kaldi-librispeech
TIMIT Unified Streaming & Non-streaming Two-pass u2-timit
Alignment THCHS30 MFA mfa-thchs30
Language Model Ngram Language Model kenlm
Speech Translation (English to Chinese) TED En-Zh Transformer + ASR MTL transformer-ted
FAT + Transformer + ASR MTL fat-st-ted

Text-to-Speech in PaddleSpeech mainly contains three modules: Text Frontend, Acoustic Model and Vocoder. Acoustic Model and Vocoder models are listed as follow:

Text-to-Speech Module Type Model Type Dataset Example
Text Frontend tn / g2p
Acoustic Model Tacotron2 LJSpeech / CSMSC tacotron2-ljspeech / tacotron2-csmsc
Transformer TTS LJSpeech transformer-ljspeech
SpeedySpeech CSMSC speedyspeech-csmsc
FastSpeech2 LJSpeech / VCTK / CSMSC / AISHELL-3 / ZH_EN / finetune fastspeech2-ljspeech / fastspeech2-vctk / fastspeech2-csmsc / fastspeech2-aishell3 / fastspeech2-zh_en / fastspeech2-finetune
DiffSinger Opencpop DiffSinger-opencpop
Vocoder WaveFlow LJSpeech waveflow-ljspeech
Parallel WaveGAN LJSpeech / VCTK / CSMSC / AISHELL-3 / Opencpop PWGAN-ljspeech / PWGAN-vctk / PWGAN-csmsc / PWGAN-aishell3 / PWGAN-opencpop
Multi Band MelGAN CSMSC Multi Band MelGAN-csmsc
Style MelGAN CSMSC Style MelGAN-csmsc
HiFiGAN LJSpeech / VCTK / CSMSC / AISHELL-3 / Opencpop HiFiGAN-ljspeech / HiFiGAN-vctk / HiFiGAN-csmsc / HiFiGAN-aishell3 / HiFiGAN-opencpop
WaveRNN CSMSC WaveRNN-csmsc
Voice Cloning GE2E Librispeech, etc. GE2E
SV2TTS (GE2E + Tacotron2) AISHELL-3 VC0
SV2TTS (GE2E + FastSpeech2) AISHELL-3 VC1
End-to-End VITS CSMSC / AISHELL-3 VITS-csmsc / VITS-aishell3

Audio Classification

Task Dataset Model Type Example
Audio Classification ESC-50 PANN pann-esc50

Keyword Spotting

Task Dataset Model Type Example
Keyword Spotting hey-snips MDTC mdtc-hey-snips

Speaker Verification

Task Dataset Model Type Example
Speaker Verification VoxCeleb1/2 ECAPA-TDNN ecapa-tdnn-voxceleb12

Speaker Diarization

Task Dataset Model Type Example
Speaker Diarization AMI ECAPA-TDNN + AHC / SC ecapa-tdnn-ami

Punctuation Restoration

Task Dataset Model Type Example
Punctuation Restoration IWLST2012_zh Ernie Linear iwslt2012-punc0


Normally, Speech SoTA, Audio SoTA and Music SoTA give you an overview of the hot academic topics in the related area. To focus on the tasks in PaddleSpeech, you will find the following guidelines are helpful to grasp the core ideas.

The Text-to-Speech module is originally called Parakeet, and now merged with this repository. If you are interested in academic research about this task, please see TTS research overview. Also, this document is a good guideline for the pipeline components.

⭐ Examples

  • PaddleBoBo: Use PaddleSpeech TTS to generate virtual human voice.


To cite PaddleSpeech for research, please use the following format.

    title = {PaddleSpeech: An Easy-to-Use All-in-One Speech Toolkit},
    author = {Hui Zhang, Tian Yuan, Junkun Chen, Xintong Li, Renjie Zheng, Yuxin Huang, Xiaojie Chen, Enlei Gong, Zeyu Chen, Xiaoguang Hu, dianhai yu, Yanjun Ma, Liang Huang},
    booktitle = {Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Demonstrations},
    year = {2022},
    publisher = {Association for Computational Linguistics},

  title = {{A}$^3${T}: Alignment-Aware Acoustic and Text Pretraining for Speech Synthesis and Editing},
  author = {Bai, He and Zheng, Renjie and Chen, Junkun and Ma, Mingbo and Li, Xintong and Huang, Liang},
  booktitle = {Proceedings of the 39th International Conference on Machine Learning},
  pages = {1399--1411},
  year = {2022},
  volume = {162},
  series = {Proceedings of Machine Learning Research},
  month = {17--23 Jul},
  publisher = {PMLR},
  pdf = {},
  url = {},

  title={Fused acoustic and text encoding for multimodal bilingual pretraining and speech translation},
  author={Zheng, Renjie and Chen, Junkun and Ma, Mingbo and Huang, Liang},
  booktitle={International Conference on Machine Learning},

Contribute to PaddleSpeech

You are warmly welcome to submit questions in discussions and bug reports in issues! Also, we highly appreciate if you are willing to contribute to this project!




PaddleSpeech is provided under the Apache-2.0 License.

Stargazers over time

Stargazers over time

Open Source Agenda is not affiliated with "PaddlePaddle DeepSpeech" Project. README Source: PaddlePaddle/PaddleSpeech