STM32 Model Zoo Help

[jzzunn]

I have followed the model zoo github tutorial to train, quantize and deploy an Audio Event Detection model. I used the available yamnet-256 and followed the tutorial to the end of deployment. I manage to get successful deployment but when I connect the B-U585I-IOT02A board, and open putty and connect via serial, i get this on putty:

Network informations...
model name : network
model signature : 0x80e2ec0fd41896ad70099555488edefd
model datetime : 2025-03-12T18:36:47+0800
compile datetime : Mar 12 2025 18:37:57
runtime version : 10.0.0
tools version : 2.0.0
complexity : 23932020 MACC
c-nodes : 21
map_activations : 1
[0] (1,1,1,112196)112196/u8 Q8.0 @0x20017480/112196
map_weights : 1
[0] (1,1,1,139176)139176/u8 Q8.0 @0x80201E0/139176
n_inputs/n_outputs : 1/1
I[0] (1,64,96,1)6144/i8 1:(0.053046,44), @0x20028180/6144
O[0] (1,1,1,10)10/float32 @0x2001748C/40

--- Start Processing ---

It stops at start processing, when I play the audio files from ESC-50 dataset, nothing happens. Am I supposed to perform some configurations on the stm32ide like enable the microphone or USART or something?

 

[Laurent]

Hello

It should normally works out of the box without additional configuration with microphone  or USART .

I think you should tune values of silence threshold located in Projects\Common\dpu\ai_model_config.h :

#define CTRL_X_CUBE_AI_SPECTROGRAM_SILENCE_THR   (123) // 0 means disabled

for info I tuned mine at 123, but you set it to zero in case you want to disable it 

I hope this can fix

best regards

L

 

 

 

[jzzunn]

I tried setting it to zero and still getting the same results. But during the deployment in the logs of command prompt i always see this error is it an issue?

Finished building: default.size.stdout

Finished building: b_u585i_iot2a_gs_aed.list

12:13:05 Build Failed. 7 errors, 8 warnings. (took 9s.656ms)

[Laurent]

Hello,

This setting should be applied on the deployed project, on the specified file, to be re-build afterwards with cube Ide  ,this done after deployment and thus should not influence the deployement itself 

 

best regards

 

L 

[jzzunn]

I notice also that when i deploy the model, it automatically sets it back to this value:

#define CTRL_X_CUBE_AI_SPECTROGRAM_SILENCE_THR   (3000.0F) // 0 means disabled

[Laurent]

Yes, and this is why it is still npt working, two solutions 

1 - you apply the fix I described after the deployment. but you need to clean, re-build and flash the project.

2 - you fix the deployment script so that the correct value is generated in the first place

in  \stm32ai-modelzoo-services\audio_event_detection\src\utils\gen_h_file.py, line 140

 if config.feature_extraction.to_db:
 f.write('#define CTRL_X_CUBE_AI_SPECTROGRAM_SILENCE_THR (0) // 0 means disabled\n')
 else:
 f.write('#define CTRL_X_CUBE_AI_SPECTROGRAM_SILENCE_THR (100.0F) // 0 means disabled\n')

I hope this help

best regards

L