Detection of nasals using a production model
K. Shirai, Tomoyuki Gomi · The Journal of the Acoustical Society of America · 1978
An important task in speech recognition is to detect nasal or nasalized sounds. In this paper, this problem is considered in the framework of an articulatory model. In this model, the position of the articulatory mechanism is expressed by six parameters containing a nasalization parameter XN which expresses the position of the velum. The transfer function of the vocal tract with a side branch was obtained from Kelly's ladder equivalent circuit. Two kinds of production models were prepared depending on whether the nasal output was dominant or not when compared with the oral output. The estimation was performed under criteria which minimized the spectral error. Isolated nasalized vowels and continuous speech including nasal consonants were analyzed. In both cases, the nasalization parameter XN was proved to be useful for the detection of nasal sounds. While all the articulatory parameters can be estimated rather successfully in the case of nasalized vowels, the method is not sufficient for discriminating /m/ and /n/ completely in continuous speech.