First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Adaptive Speech Processing Using an Accelerated Stochastic Approximation Method

Hirofumi Yogo, Naoki Inagaki

Department of Electrical and Computer Engineering, Nagoya Institute of Technology, Nagoya, Japan

This paper presents a speech parameter estimation based on adaptive digital filtration(ADF). Accelerated stochastic approximation methodology with improved convergence property and parameter estimation accuracy is used as an adaptive algorithm for ADF. ADF has a performance which converges within 25 iterations even if the input signal has a considerablly higher autocorrelation;for example,a correlation coefficient of 0.99. Parameter estimation,therefore, within a one pitch period is easily performed,and the method can process both male and female speech without any restrictions. This paper describes the application of a method to parameter estimation for stop consonants and vowels. The identification of the stop consonants by neural network is described.

Full Paper

Bibliographic reference.  Yogo, Hirofumi / Inagaki, Naoki (1990): "Adaptive speech processing using an accelerated stochastic approximation method", In ICSLP-1990, 481-484.