echosonic demonstrates the capability of a physical reservoir computing system to provide computational power for a sounds classification task. The demonstration consists in an urban sounds recognition task of a standard dataset using a non-linear oscillator as a reservoir computer imitating a deep digital neural networks to carry out audio recognition.
What happen: The input audio signal is mixed with an sinusoidal signal and converted to an analog signal to be sent to the computational circuit. The outputs from the circuit are collected at a rate which will determine the number of virtual nodes of the network. Once the features are collected, a simple optimization algorithm is use to identify the class.
Comments