Technical Report of IEICE Japan, Vol. SP2002-64, pp. 41-46, 2002 (in Japanese)
Continuous speech recognition based on the contribution of modulation frequency components
N. Kanedera, T. Arai, K. Okada and Y. Momomura
Abstract: The Fourier transform of the time trajectories of a parameter such as logarithmic spectrum or cepstrum is called the modulation spectrum. In this paper we propose new feature for automatic speech recognition based on contribution of modulation frequency components. The contribution shows the importance of each modulation frequency component for speech recognition. Testing proposed feature on IPA98 task in noisy environments (SNR10 dB) gave a relative improvement of 5% in word accuracy over the MFCC with dynamic feature.
Keywords: feature for automatic speech recognition, modulation spectrum, modulation frequency, noise