First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Early Developments of LPC Speech Coding Techniques

Fumitada Itakura

Department of Electrical Engineering, Nagoya University, Nagoya, Japan

Since the invention of a speech analysis-synthesis system based on the maximum likelihood(ML) spectrum estimation in 1966 by Shuzo Saito and the author, research and development of efficient speech coding methods and their applications to computer and communication systems have been actively conducted in Japan. Our interest in LPC speech coding goes back to 1965 when I was looking for new research project for the graduate study at Nagoya University, after finishing the master degree on time series analysis of the bio-medical signals. At that time, an excellent book "Speech Analysis Synthesis and Perception" by Jim Flanagan was just printed and a review paper on vocoders by Manfred Schroeder was published in a short time later. Previous efforts to encode speech signal were well summarized in these documents. After examining these, I was convinced that the most fundamental problem in speech processing to achieve efficient speech coding and speech recognition, is to extract the spectral information from speech waveform as accurately and efficiently as possible. Accordingly, I tried to attack the problem from the view point of a statistical analysis of stationary random processes. The first outcome was the maximum likelihood (ML) spectral estimation of the all-pole spectrum. This method was motivated by the profound results on the hypothesis testing in time series analysis by Peter Whittle. The second outcome was the partial autocorrelation (PARCOR) analysis. The statistial concept of the partial autocorrelation was introduced to speech analysis, and the resulting lattice structure for the analysis and synthesis filter has now been widely used for general speech processing. The third development is the line spectrum pair(LSP) representation of speech spectrum. The concept of LSP was motivated by the work of Ya. L. Geronimus on the orthogonal polynomials on a circle (1948).

In this talk, I would like to present a historical retrospective of the early developments of LPC speech analysis and coding techniques and their impacts on the recent developments of new generation speech coding methods and speech recognition.

Full Paper

Bibliographic reference.  Itakura, Fumitada (1990): "Early developments of LPC speech coding techniques", In ICSLP-1990, 1409-1410.