support for xiao esp32s3 sense (AIS-1464)

[gparvind]

I tried configuring wake word detection example for xiao esp32s3 sense but it does't detect any spoken wake words.
I have set the config for pdm mono mode single mic config. Tried with sample rate 441000 and 16000 both not working. When I print feed data, I can it change when spoken something in mic. With same mic i2s configuration, esp-idf i2s_recorder exanple is working.
Anything specifig I am missing?

[gparvind]

vad_state shows speech activity so microphone is getting detected but not the wake words.

[gparvind]

@feizi any help with this?

[sun-xiangyu]

Can you save audio data of your board by WiFi or SD card? I need audio data to see if the recording is correct.
The sample rate is 16KHz.

[wzdl]

Hi, I'm also trying to run wake word detection on the Xiao ESP32-S3 Sense board. I've encountered similar issues, but I managed to get it working by modifying the pcm_config .

afe_config.pcm_config.total_ch_num = 1;
afe_config.pcm_config.mic_num = 1;
afe_config.pcm_config.ref_num = 0;


However, even though I've loaded multiple models, only the wakenet9l_tts1h8_你好小鑫 model seems to work, and the recognition accuracy is not satisfactory.

Are there any methods to improve recognition accuracy?

[lixy123]

https://blog.csdn.net/weixin_44135213/article/details/137217731
There are some clues. Actual measurement cannot be used. it still cannot be awakened. Further efforts are needed.