First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Pole-Zero Structure Based on Two-Source Vocal Tract Model, PSE Inspection of Continuous Speech Vowel Part

Takayuki Nakajima, Hiroshi Ohmura

Electrotechnical Laboratory, Tsukuba Science City, Japan

In this paper, at first, Japanese continuous speech vowel parts PSE (power spectrum envelope) are ob- served, and it is pointed out that there is a great difference between the continuous speech vowel PSE and the.PSE of a vowel uttered in isolation. In order to explain the phenomena, a two-source vocal tract model is proposed. In the model, the vocal tract is expressed as a non-uniform single tube including the under glottal portion. Giving the vowel vocal tract area function corresponding to the 5 vowels of Japanese and two other parameters: the source position and the power ratio, vocal tract pole-zero transfer functions are calculated and an articulatory/acoustic nomogram is constructed. As the results of the inspection if there is a correspondence between the PSE of real speech and one of the nomogram's frequency transfer characteristics in pole-zeroes, the authors present the following hypothesis: "The continuous speech vowel part PSE has the kind of pole-zero structure generated by the two-source vocal tract model".

Full Paper

Bibliographic reference.  Nakajima, Takayuki / Ohmura, Hiroshi (1990): "Pole-zero structure based on two-source vocal tract model, PSE inspection of continuous speech vowel part", In ICSLP-1990, 177-180.