A discrete wavelet transform based approach to Hindi speech recognition

In this paper, we propose a new scheme for recognition of isolated words in Hindi Language speech, based on the Discrete Wavelet Transform. We first compute the Discrete Wavelet Transform coefficients of the speech signal. Then, Linear Predictive Coding Coefficients of the Discrete Wavelet Transform coefficients are calculated. Our scheme then uses K Means Algorithm on the obtained Linear Predictive Coding Coefficients to form a Vector Quantized codebook. Recognition of a spoken Hindi word is carried out by first calculating its Discrete Wavelet Transform Coefficients, followed by Linear Predictive Coding Coefficient calculation of these Discrete Wavelet Transform Coefficients, and then deciding in favor of the Hindi word whose corresponding centroid (in the Vector Quantized codebook) gives a minimum squared Euclidean distance error with respect to the word under test.