ESP-SR Speech Recognition Framework

May 25, 2026 · View on GitHub

Documentation Status Component Registry

Espressif ESP-SR helps users build AI speech solutions.

Overview

ESP-SR framework includes the following modules:

These algorithms are provided in the form of a component, so they can be integrated into your projects with minimum effort.

News

  • [2026/5/09]: Preliminary support for ESP32-S31 target.
  • [2026/4/28]: We update a new AEC algorithm for full-duplex scenarios. For performance evaluation and usage instructions, please refer to the documentation.
  • [2026/4/23]: TTS Pipeline V3 now supports wake word training for Chinese, English, Japanese, and French. Planned support includes Korean, Spanish, Portuguese, German, Russian, and Arabic.
  • [2025/10/20]: We add a new model, WakeNet9l, which further improves the response rate of wake words spoken at extremely fast speeds based on WakeNet9. The usage of WakeNet9l is the same as WakeNet9, but its CPU and memory requirements are approximately 1.3 times higher than those of WakeNet9.
  • [2025/4/21]: We add a new model WakeNet9s, which can run on chips that do not have PSRAM and do not support SIMD, such as ESP32C3 and ESP32C5. examples
  • [2025/4/17]: We add a new DOA(Direction of Arrival) algorithm.
  • [2025/2/14]: We release ESP-SR V2.0. Migration from ESP-SR V1.* to ESP-SR V2.*
  • [2025/2/13]: We release VADNet, a voice activity detection model. You can use it to replace the WebRTC VAD and improve the performance.

Wake Word Engine

Supported TargetsESP32ESP32-S2ESP32-S3ESP32-S31ESP32-P4ESP32-C3ESP32-C5ESP32-C6

Espressif wake word engine WakeNet is specially designed to provide a high performance and low memory footprint wake word detection algorithm for users, which enables devices always listen to wake words, such as “Alexa”, “Hi,lexin” and “Hi,ESP”. WakeNet9 and WakeNet9s models are supported. WakeNet9s is a cost-down version of WakeNet9, with fewer parameters and lower computational requirements. _tts suffix means this WakeNet model is trained by TTS samples. _tts2 suffix means this WakeNet model is trained by TTS Pipeline V2. TTS Pipeline V3 start to support more language.

Espressif offers two ways to customize the wake word, please refer to the following document to choose the one that meets your needs: Espressif Speech Wake Words Customization Process or Training Wake Words by TTS sample.

The following wake words are supported in esp-sr:

wake wordsWakeNet9sWakeNet9
Hi,乐鑫wn9s_hilexinwn9_hilexin
Hi,ESPwn9s_hiespwn9_hiesp
こんにちは ESPwn9l_ja_konnichihaesp_tts3
Bonjour ESPwn9l_fr_bonjouresp_tts3
你好小智wn9s_nihaoxiaozhiwn9_nihaoxiaozhi_tts
Hi,Jasonwn9s_hijason_tts2wn9_hijason_tts2
你好喵伴wn9_nihaomiaoban_tts2
小爱同学wn9_xiaoaitongxue
Hi,M Fivewn9_himfive
Alexawn9_alexa
Jarviswn9_jarvis_tts
Computerwn9_computer_tts
Hey,Willowwn9_heywillow_tts
Sophiawn9_sophia_tts
Mycroftwn9_mycroft_tts
Hey,Printerwn9_heyprinter_tts
Hi,Joywn9_hijoy_tts
Hey,Wandwn9_heywanda_tts
Astrolabewn9_astrolabe_tts
Hey,Ilywn9_heyily_tts2
Hi,Jollywn9_hijolly_tts2
Hi,Fairywn9_hifairy_tts2
Blue Chipwn9_bluechip_tts2
Hi,Andywn9_hiandy_tts2
Hey,Ivywn9_heyivy_tts2
Hi,Stack Chanwn9l_histackchan_tts3
Hey,Kirawn9_heykira_tts3
Hi,Wall E/Hi,瓦力wn9_hiwalle_tts2
你好小鑫wn9_nihaoxiaoxin_tts
小美同学wn9_xiaomeitongxue_tts
Hi,小星wn9_hixiaoxing_tts
小龙小龙wn9_xiaolongxiaolong_tts
喵喵同学wn9_miaomiaotongxue_tts
Hi,喵喵wn9_himiaomiao_tts
Hi,Lily/Hi,莉莉wn9_hilili_tts
Hi,Telly/Hi,泰力wn9_hitelly_tts
小滨小滨/小冰小冰wn9_xiaobinxiaobin_tts
Hi,小巫wn9_haixiaowu_tts
小鸭小鸭wn9_xiaoyaxiaoya_tts2
璃奈板wn9_linaiban_tts2
小酥肉wn9_xiaosurou_tts2
小宇同学wn9_xiaoyutongxue_tts2
小明同学wn9_xiaomingtongxue_tts2
小康同学wn9_xiaokangtongxue_tts2
小箭小箭wn9_xiaojianxiaojian_tts2
小特小特wn9_xiaotexiaote_tts2
你好小益wn9_nihaoxiaoyi_tts2
你好百应wn9_nihaobaiying_tts2
小鹿小鹿wn9_xiaoluxiaolu_tts2
你好东东wn9_nihaodongdong_tts2
你好小安wn9_nihaoxiaoan_tts2
你好小脉wn9_ni3hao3xiao3mai4_tts2
你好小瑞wn9_ni3hao3xiao3rui4_tts3
嗨小欧wn9_hai1xiao3ou1_tts3
小珈小珈wn9_xiao3jia1xiao3jia1_tts3
小峰小峰wn9_xiao3feng1xiao3feng1_tts3
嗨小象wn9_hai1xiao3xiang4_tts3
你好星宝wn9l_ni3hao3xing1bao3_tts3

NOTE:

The product names, logos, and brands associated with the wake words listed in this software are the property of their respective owners. They are shown here solely to provide examples of wake words for users to understand and test Espressif’s Speech Recognition Framework, and do not imply any affiliation with or endorsement by their owners. This software is not affiliated with, endorsed by, or in any way officially connected to any trademark owner. Before any commercial use, you must ensure that you are the lawful rights holder of the relevant wake words or have obtained proper authorization from the lawful rights holder.

本软件所列唤醒词相关的产品名称、标识及品牌均归其各自所有权人所有。展示这些内容仅为了提供唤醒词示例，供用户了解和测试乐鑫提供的语音识别框架，并不表示与权利人存在任何关联或获得认可。本软件与任何商标权人无官方关系。在任何商业使用前，请确保您拥有相关唤醒词的合法权利或已获得合法授权。

Speech Command Recognition

Supported TargetsESP32ESP32-S3ESP32-P4ESP32-S31

Espressif's speech command recognition model MultiNet is specially designed to provide a flexible off-line speech command recognition model. With this model, you can easily add your own speech commands, eliminating the need to train model again.

Currently, Espressif MultiNet supports up to 300 Chinese or English speech commands, such as “打开空调” (Turn on the air conditioner) and “打开卧室灯” (Turn on the bedroom light).

The following MultiNet models are supported in esp-sr:

languageESP32ESP32-S3ESP32-P4/ESP32-S31
Chinesemn2_cnmn5q8_cn, mn6_cn, mn7_cnmn7_cn
Englishmn5q8_en, mn6_en, mn7_enmn7_en

Audio Front End

Supported TargetsESP32ESP32-S3ESP32-P4ESP32-S31

Espressif Audio Front-End AFE integrates AEC (Acoustic Echo Cancellation), VAD (Voice Activity Detection), BSS (Blind Source Separation) and NS (Noise Suppression), NSNET(Deep noise suppression) and other functions. It is designed to be used with the ESP-SR library.

Our two-mic Audio Front-End (AFE) have been qualified as a “Software Audio Front-End Solution” for Amazon Alexa Built-in devices.

Documentation and Resources