ESP32-S3 has AI instructions and faster PSRAM. In general， the performance of ESP32-S3 on AI is 5~10 times that of ESP32. The parameters of new English Multinet model have exceeded 2M. It is too big to run in real time on ESP32. We strongly recommend that you use the ESP32-S3 board to run AI-related...

We have updated a new "Hi,ESP" model by adding more training samples.
You can pull the latest esp-sr (release/v1.0) to try it. https://github.com/espressif/esp-sr/tree/release/v1.0

Hi Pibbotley
I have checked the firmware v0.2.1 for the ESP-BOX. We can wake up by "Hi,ESP"
When I play the video https://youtu.be/n0jH1Brmee8, the ESP-BOX can not be woken up.
Can you try the attached sample to see if the sample can work.

最近我们esp-sr更新的比较频繁，请问你使用ADF中的esp-sr的commit是多少
如果可以，请你试一下 esp-skainet下面的example/cn_speech_commands_recognition
https://github.com/espressif/esp-skaine ... ecognition

目前在esp-sr中
esp32s2只支持 中文TTS，（esp32s2缺乏一些必要的加速指令，所以我们暂不支持所有example）
esp32支持所有example，
esp32s3也将支持所有example，预计在9月初release

MultiNet是支持编译后重新配置唤醒词，但是目前并没有开发相关的API，后续会提供支持