How can I perform real time predictions on MCU (NucleoL496ZG-P) based on my classificator keras model?

[NŠlju.1]

Hello, I am new to deployment of neural networks on mcu's. I am doing a speech enhancment project and want to deploy my noise classificator on my nucleo L496ZGP board. I have 0 exprience with this, how can I do tests on my custom data and (prefferebly) classify recorded sounds in real time in xcube-ai?

[fauvarque.daniel]

A first step is to look at the examples in the embedded documentation, in STM32CubeMX once you have a project with X-Cube-AI added just do Help->X-Cube-AI Documentation to access the documentation.

In STM32CubeMX you can also generate a template application that will give you an example on how to use the API, just select Application Template as the additional component of the pack.

The template code is generated under app_x-cube-ai.c

Regards

[Laurent FOLLIOT · Accepted solution]

Hello,

You could also use the ST model zoo : https://github.com/STMicroelectronics/stm32ai-modelzoo/tree/main/audio_event_detection

We provided an example to train, evaluate and deploy an AED model on U585 device (not L496 device but this may be of interest anyway).

Regards

[NŠlju.1]

Thank you!