Audio-classification-with-transfer-learning

This project was developed to classify and identify the 'Normal' and 'Reading' mode of audio sounds. The main task of this project was to develop a prediction tool that can classify between the both classes as either normal or reading. Inspired from the prior works on similar sound classification, we developed a system that uses a set of features extracted from spectral features of sounds like pitch, fundamental frequency, first and second order sound frequency to train a neural network model over a VGG19 pre-trained model (transfer-learning).

#MFCCs-FEATURES

MFCCs are widely used in SVD (singing voice detection) and were first introduced by Davis and Mermelstein in 1980. The use of MFCCs has proven to be a powerful tool in music and voice recognition, and sound recognition in general. (link: https://encyclopedia.pub/entry/18717)

The MFCCs are calculated as follows:

1. Division of the speech signals into frames, usually by applying a windowing function at fixed intervals;
2. Computing the coefficients of the discrete Fourier transform on each segment of windowed signal to convert the time domain into the frequency domain;
3. Taking the logarithm of the amplitude spectrum;
4. Smoothing the spectrum and emphasizing perceptually meaningful frequencies ;
5. Taking the discrete cosine transform (DCT) of the list of mel log powers;
6. Generating cepstrum.