Sound Recognition with Deep Learning Methods

The aim of the thesis is to design a communication system based encoding-decoding and encrypted sound code with Forward Error Correction and Short-Time Fourier Transform and using forward error correction algorithm is Reed-solomon and scrambling methods . Using sound to recognize transportation mode and propose a deep-learning CNN network operating on the STFT log spectrum of the sound. With recognition app using Tenserflow and evaluating MEL diagrams and comparing with Confusion Matrices to obtain the best frequency for the aim of the project.