An efficient and simple approach to segment the consonant and the vowel (C/V) speech signals by slightly corrupting a speech signal with white noise is proposed. This approach is based on the vowels are quasi-periodic, and the consonants can be regarded as broad-spectrum noises. A Fourier-based speech enhancement algorithm performs well in enhancing the corrupted vowel signals, and this algorithm fails in enhancing the corrupted consonant signals.
Experimental results show that the novel scheme performs well in detecting the C/V boundaries for a spontaneously spoken utterance.
