Proc. of the Forum Acusticum Sevilla, 2002
Discrimination of speech from environmental noises using the modulation cepstrum
T. Doi, T. Ishida, T. Goto, T. Miyoshi, T. Arai and Y. Murahara
Abstract: We introduced “the modulation cepstrum,” a novel representation of an acoustic signal used to distinguish speech from environmental noises, the latter being a potential cause of performance degradation of speech related processing. The moduration cepstrum was computed by taking the inverse Fourier transform of the logarithmic modulation spectrum, a spectral representation of the temporal dynamics of the signal. An experiment showed that the modulation cepstrum enabled us to identify an input signal as speech when the position of its center of gravity was less than 125 ms on the modulation cepstral domain.