ESP-EYE runs glass-break-detection issue

Help
#1

Hi,

I use the https://studio.edgeimpulse.com/studio/139844 glass break detection project and deployment to arduino library. Then i add the library to Arduino and run the esp32_mircophone.ino. But i check the serial log, it always displays the “Glass_Breaking: 0.996094” not as background. i do not know if the issue from edge impulse development or arduino esp32_microphone.ino. Appreciate if anyone could help.

Predictions (DSP: 506 ms., Classification: 12 ms., Anomaly: 0 ms.): 
    Background: 0.000000
    Glass_Breaking: 0.996094
Predictions (DSP: 507 ms., Classification: 12 ms., Anomaly: 0 ms.): 
    Background: 0.000000
    Glass_Breaking: 0.996094
Predictions (DSP: 506 ms., Classification: 12 ms., Anomaly: 0 ms.): 
    Background: 0.000000
    Glass_Breaking: 0.996094
Predictions (DSP: 507 ms., Classification: 12 ms., Anomaly: 0 ms.): 
    Background: 0.000000
    Glass_Breaking: 0.996094
#2

Hi @xyu6,

To start troubleshooting, I recommend checking to make sure that your microphone works. If you load the pre-compiled ESP-EYE firmware onto Edge Impulse, can you connect the board to your Edge Impulse project via WebUSB and record a few samples? Play those samples in Studio to see if the sound was recorded correctly.

Next, how did you collect the data of the glass breaking? I have found that the microphone can sometimes make a difference. If you collect data with one microphone and then deploy to a device with a different microphone, that can often affect the inference results.

Finally, try the static_buffer example in the Arduino library downloaded from your project. Try copying in the raw features from a test sample (i.e. in your project) to see if the ESP-EYE gives you close to the same inference result as in your Edge Impulse project.