On the development of speech technology

According to the methodology of signal processing the history of speech signal processing was divided into three stages:waveform processing, parameter processing, and information processing. Both source-filter model and Hidden Markov Model have constructed the theoretical bases and supplied the powerful tools. However the products of man-machine speech communication systems based on current speech technology are still not satisfactory in practical applications. Some limitations of these models and troubles in using them were discussed in this paper. It is emphasized that the diagnostic tests have to be added to the performance evaluation of speech recognition systems and speech synthesis systems, in order to improve the operation of these models and to develop new models. The theory of "more data is better data" that seems like force-fed duck style education can not be the way out. And now the speech technology has come to a standstill, more and more fundamental researches are needed to achieve a new stage of speech technology. The real motive power of development of speech technology comes from both market pull and science and technology push.