Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Glottal Excitation Modeling Using HMM with Application to Robust Analysis of Speech Signal

Akira Sasou, Kazuyo Tanaka

Electrotechnical Laboratory, AIST, MITI, Tsukuba, Ibaraki, Japan

This paper describes a robust analysis method for high fundamental frequency speech signal. In the proposed method, a Hidden Markov Model (HMM) is applied in order to represent the non-stationary property of the glottal source. Experiments are carried out using both synthetic and natural speeches to confirm the effectiveness of the method. Experimental results indicate (1) in the case of using synthetic speech in the pitch range of up to 750Hz, the proposed method can precisely estimate the original spectrum, and (2) the spectrum estimated from natural speech of pitch frequency 666Hz is less affected by the harmonics of glottal excitation, compared with result of the conventional method.

Full Paper

Bibliographic reference.  Sasou, Akira / Tanaka, Kazuyo (2000): "Glottal excitation modeling using HMM with application to robust analysis of speech signal", In ICSLP-2000, vol.4, 704-707.