Whistle Detection

Our whistle detection algorithm is based on a deep learning algorithm which makes use of the Fourier Transform. Based on the prediction on each ear sensor, the final prediction is made.

Fourier Transform

The input is taken continuously till a whistle is heard by the robot. A short term Fourier transform of the input is taken which converts the audio input to a spectrogram. Whistles usually have high pitches and thus a characteristic spectrogram which is then identified by the deep learning model.

Spectrogram classification

A deep learning model is built using Tensorflow and it takes the spectrogram and predicts whether it is a whistle or not. To make the method more robust, we use the audio input from both the ears. If the confidence level exceeds a particular threshold on both the ears, the robot sends a signal that a whistle has been blown.