Figure 9. Stblesensor - Baby Crying Detection Process; Figure 10. Stblesensor - Baby Crying App Neural Network - ST STEVAL-MKSBOX1V1 User Manual

UM2580
Mode 1 example apps
The FFT feature extraction of the acquired signal is processed by the STM32 MCU which calculates the MEL FFT
and the MFCC (MEL frequency cepstral coefficient) parameters sent to the implemented MCU neural network: if a
baby crying event is detected, the green user LED on the Sensortile.box board lights up and a warning is sent to
the smartphone via Bluetooth.
Figure 9. STBLESensor - baby crying detection process
The neural network is classified as a deep feed forward neural network and its structure is composed of 2 hidden
nodes of 100 neurons each.
The tool used to develop the neural network is Keras with an open source high level library written in Python.
Optimization and loading of the neural network on the Sensortile.box has been performed using
STM32CubeMX.AI.
Figure 10. STBLESensor - baby crying app neural network
UM2580 - Rev 5
page 8/41