Training model on MXRT1060-EVK

[nahan_trogn]

Hello,

Can I train a model on my MIMXRT1060-EVK devkit?

I ask this question because when I bring my tflite model (yamnet, recognize acoustic) from PC to my devkit, the model work well with the static array audio, it  can detect right lable; When I use it to recognize real-time audio, it detect very wrong label.

Then, I went back to my PC and change a new micro, the model on my PC fails to get proper label. So I think that the problem come from the Micro, different micro affect to my model's precision.

I want to use the micro on my dev-board to retrain my model, can I train model on it?

[kerryzhou]

Hi nahan_trogn,

   As I know, the model is trained in the PC side, not the MCU side.

  Do you try to save the real audio data to the memory, then recognize it, whether it is correct or not?

  Do you use the RT1060 SDK eiq on your side?

  If you still have questions about it, could you please give me the detail reproduce steps and your related code? I would like to help you to check it on my side.

Best Regards,

Kerry