Fast Dynamic Speech Recognition via Discrete Tchebichef Transform

Traditionally, speech recognition requires large computational windows. This paper proposes an approach based on 256 discrete orthonormal Tchebichef polynomials for efficient speech recognition. The method uses a simplified set of recurrence relation matrix to compute within each window. Unlike the Fast Fourier Transform (FFT), discrete orthonormal Tchebichef transform (DTT) provides simpler matrix setting which involves real coefficient number only. The comparison among 256 DTT, 1024 DTT and 1024 FFT has been done to recognize five vowels and five consonants. The experimental results show the practical advantage of 256 Discrete Tchebichef Transform in term of spectral frequency and time taken of speech recognition performance. 256 DTT produces frequency formants relatively identical similar output with 1024 DTT and 1024 FFT in term of speech recognition. The 256 DTT has a potential to be a competitive candidate for computationally efficient dynamic speech recognition.