Automated speech recognition system with noise consideration [articol]

Abstract

In the real world environment, Speech recognition is one of the most progressing research area. The quality as well as performance of automated speech recognition is disturbed by noises exist in the speech signals. Noises are inevitable in the speech that is transferred via external medium containing noise. In the previous research, it is resolved with the help of Multivariate Autoregressive Spectrogram Modeling (MARSM). On the other hand, in the previous method, noise reduction in the speech recognition is carried out by means of taking the greater energy coefficients as well as the signal with greater correlation. By concentrating on these, it is presumed that the noise existence is evaded significantly. In the research system, it is solved by means of presenting the new technique known as Background Noise concerned Automated Speech Recognition System (BN-ASR).In the presented system, Noise reduction is carried out with the help of the Normalized data nonlinearity (NDN)-LMS adaptation technique. This technique could adaptively remove the noises exist in the signals. Subsequent to noise reduction feature extraction is carried out with the help of the technique called Synchrony-Based Feature Extraction that will foresee the averaged localized synchrony response of the noise filter. At last, for the precise recognition of speech signals, Hybrid Particle Swarm Optimization-Artificial Neural Network (HPSO-ANN) is introduced. In the matlab simulation environment, the complete implementation of the research method is performed and it is clear that the research technique results in providing the best possible outcome compared to the previous research techniques.