Automatic Speech Recognition System Based on Hybrid Feature Extraction Techniques Using TEO-PWP for in Real Noisy Environment

Automatic speech recognition presents an interesting research area that has always attracted researchers to the general public. It is now giving rise to an important set of applications of a very varied nature and difficulty, involving millions of people around the world every day. In this paper, a model of speech recognition system in noisy environment is developed and analyzed. The proposed model relies on several hybrid feature extraction methods. Indeed, Teager-Energy Operator, Perceptual Wavelet Packet (TEO-PWP), Mel Cepstrum Coefficient (MFCC) and Perceptual Linear Production (PLP) are combined to construct a robust HMM based system. TIMIT database, which consist of both clean and noisy speech files recorded at different level of Speech-to-Noise Ratio (SNR, has been used for the system test. Results and observations are performed to prove the effectiveness of the proposed system relying on speech recognition rates.