Clustering Approach in Speech Phoneme Recognition Based on Statistical Analysis

Abstract

In general, speech recognition is a process that is referred to convert spoken string into machine-understandable string. Speech Recognition consists of 2 processes, i) removal of background noise (background noise is generated due to the stressful noise environment) and ii) phoneme separation word by word (also involves phoneme recognition). In real time situation, sound signals consist of both noises (target noise as well as background noise).This paper critically evaluates the currently available signal analysis techniques and the modeling of phonemes, as applied to isolated and context-independent phoneme recognition. The proposed methodology introduces the technique of determining the pure speech-signal in a noisy environment (without background noise) and phonemes-isolation word by word using some clustering approach. With the use of proposed methodology, high accuracy of background noise-isolation (obtaining clean speech-signal without background noise) and high accuracy of phoneme isolation from clean speech-signal have been achieved which can be qualitatively compared to previous research done on continuous phoneme recognition. Performance evaluation also shows the improvement to achieve the speech recognition in a stressful noise situation and better quality of phoneme separation process.KeywordsPhonemeg2pTTSICACASA