Acoustic features of nasal consonants in VC contexts

Abstract

The acoustic features of the nasals [m,n] in CV syllables and [m,n,ŋ] in VC syllables in English were first analyzed and compared using cepstrally smoothed running FFT spectra. Clear differences between nasals in CV and VC syllables were obtained. For example, the spectral energy transitions from vowel to nasal in VC syllables were found to be much less dramatic than in CV transitions. Next, given the recent interest in the efficacy of auditory representations in speech recognition schemes, the nasals in VC contexts were examined in terms of auditory transformed running spectra. Several features of interest were obtained: The spectra were generally dominated by the second formant and, when the preceding vowel was a low vowel, nasal place of articulation was distinguished by the second formant transition, which converged to 16–17 ERB for [m] and 19–20 ERB for [n]; [ŋ] was characterized by little formant movement. Consistent nasal place features were not found in the context of the vowels [i] and [u]. Finally, since the antiresonances of a nasal may provide place of articulation information, the system zeros of the nasals were analyzed using parametric spectral analysis based on the autoregressive moving average (ARMA) process. Results indicated that the antiresonances of nasals in a VC syllable could be consistently estimated by a two‐step AR approximation method that could be used to distinguish reliably between [m] and [n].