Global SNR Estimation of Speech Signals

Sreelakshmi M.S., Ann Nita Netto

Abstract: Speech processing deals with the study of speech signals and the processing methods of these signals. The performance of these speech signals gets reduced due to several environmental conditions. The estimation of signal-to-noise ratio (SNR) gives an idea about the amount of noise present in the original signal. SNR compares the level of the data signal to the level of background noise. A novel method based on certain features of the speech signal is used for estimating the SNR. From the features a regression model is build to estimate SNR. Mel frequency cepstrum coefficients (MFCC) feature extraction technique is used for the effective removal of noise from the real-time speech signal.

Keywords: Speech processing, MFCC, Signal-to-noise ratio, DNN