A Fuzzy C-Means based GMM for Classifying Speech and Music Signals

Gaussian Mixture Model (GMM) with Fuzzy c-means attempts to classify signals into speech and music. Feature extraction is done before classification. The classification accuracy mainly relays on the strength of the feature extraction techniques. Simple audio features such as Time domain and Frequency domain are adopted. The time domain features are Zero Crossing Rate (ZCR) and Short Time Energy (STE). The frequency domain features are Spectral Centroid (SC), Spectral Flux (SF), Spectral Roll-off (SR) and Spectral Entropy (SE) and Discrete Wavelet Transforms. The features thus extracted are used for classification. Commonly GMM uses Expectation Maximization (EM) algorithm to determine parameters. The proposed GMM makes use of fuzzy c-means algorithm. The fuzzy c-means algorithm is used to estimate the parameters of the GMM. Compute the probability density function and fix the Gaussian parameter. The proposed GMM model classifies the given input signal is either speech or music and compared with GMM using EM algorithm.