>> Top >> Research plan >> Speech and language information processing group

The improvement of vocal-tract models for the analysis of the physical acoustic characteristics of speech production system, and for the application to the interface of intellectual systems

The improvement of acoustic models of the speech production system is important for the progress of the speech related technologies, such as speech synthesis, speech recognition, speaker recognition, or kansei information processing of speech. Recent development of techniques for the observation of speech organs such as MRI allows us to obtain accurate descriptions of vocal tract shape. The detailed analysis of the acoustic characteristics of the three-dimensional vocal-tract is necessary to obtain the advanced information of the speech such as naturalness or individuality. In this research project, we investigate on the fine acoustic characteristics of vocal-tract by using numerical computation techniques. The acoustic measurement is also important to obtain basic data of the flow-acoustic phenomenon which is not easy to solve by numerical simulations. Based on these results, we try to construct an acoustic speech production model that is effective not only in ordinary lower frequencies but also in higher frequencies which are important for consonants like fricatives and for the individuality of speech. The realization of a high-performance speech processing system with a realistic model of the glottal sound source and the vocal tract wall will be also investigated.

Finite element meshes of /a/. (left).
Sound-pressure distribution in vocal-tract of /a/(3654Hz). (right)

Spoken dialogue process based on learning capability for acquisition of language and knowledge

A dialogue is very important as the method of communication in our life, and it is based on language. Human beings dynamically acquire a language on the growth process. A infant cannot speak language and has no knowledge. However it grows into an adult that can speak language very well and has various kinds of knowledge by learning from the environment. Therefore a human being has native capability for acquisition of language and knowledge certainly. The final goal of our research is that "Realization of the system that can acquire various kinds of knowledge based on learning capability, and communicate by dialogue." This means that our system can adapt to each user dynamically. We plan to build the intellectual robot that helps with the housework based on spoken dialogue process as one of the application of our research.

Advanced improvements of natural language information processing and the application to intelligent language information systems

For the advanced improvements of natural language information processing, we must study about the integrated models for processing knowledge and context information and the possibilities of their implementations from the various points of view. In this research project, we consider the integrated methods for processing knowledge and context information in human discourse processing, and the machine translation system from the Ainu Language into Japanese as an implementation of natural language processing systems. Furthermore we are going to develop an intellectual and introductory learning system for the Ainu Language including the automatic translation mechanism.

>> Vision and image information processing group
>> Robot and control information processing group