An Approach for Speech Recognition via Voice Activity Detection

Research is being done on Automatic Speech Recognition (ASR), which can be used effectively in noisy environments. In terms of lustiness, the effectiveness of popular parameterization techniques was assessed in contrast to the background signal. A hybrid feature extractor is used for Mel frequency cepstral coefficients (MFCC), Perceptual linear predictive (PLP) coefficients, and their modified forms by combining the fundamental building blocks of PLP and MFCC. The VAD-based frame dropping formula was only applied to the ASR method’s training phase. This method has the advantage of removing pauses and potentially significantly distorted speech segments, which aids in more precise phone modelling. The second portion focuses on the examination and contribution of the modified vocal activity detection technique.