by Jan Uhlír
The project Theory and Application of Speech Communication in Czech supported by the Grant Agency of the Czech Republic is oriented to the theory and design of systems in the telecommunications, multimedia, and handicap help application areas. To empower disabled people, the project targets cochlear implants and their optimal application, speech enhancement for hearing impaired, voice controlled systems for blind operators (telephone switch, computer programming), text to speech (TTS) systems for blind people, and help for education and training of handicapped people.
Disability of speech communication both in the office and in the family stands the hearing impaired into the social isolation. Fortunately, it is possible to compensate for the hearing loss, up to great extent, using the modern hearing aids like the cochlear implantation, which gives hope on partial restoration of hearing also to the totally deaf. However, persons with the hearing impairment ranking from severe to total hearing loss, compensated for by a hearing aid or by a cochlear implant, come very often into situation in which they can understand the speech only with the greatest difficulties. This occurs especially in the noisy environment and in situations when several persons are speaking at the same time (cocktail party effect). Hearing aids and cochlear implant manufacturers are performing research aiming to enhancement of the speech recognition under these conditions. Some achievements has been attained in this field but the problem as a whole still remains unsolved.
The research team from the Department of Circuit Theory of the University works on a method and its device implementation which would enable to suppress the noise present in the signal taken by the microphone in a noisy environment and eliminate the cocktail party effect. Using that device should improve the communication ability of the hearing impaired.
The blind people and paraplegic disabled are targeted by the text to speech part of the project. The TTS synthesis in contradistinction to other scientific problems is language dependent and the TTS systems developed for one language can not be applied for other languages. It is the reason why many countries are developing their own systems, which would be acceptable for users, ie not only intelligible but also producing natural sounding speech.
TTS synthesis is a multidomain problem requiring co-operation of scientists in phonetics, phonology, signal and system theory and computer science. In the last five years in co-operation between the Institute of Radio Engineering and Electronics, Academy of Sciences and of the Institute of Phonetics, Faculty of Arts, Charles University a new TTS system was proposed, which is based in diphones and subphonemical units extracted from natural speech and modelled by a parametric method. This system was implemented on a signal processor and is in production in the Co-operative of the Blind SPEKTRA. It is appreciated also abroad for high intelligibility, naturalness, compactness and broad application. The main research interest in this part of the project is enhancing the naturalness of synthetic speech by more precise speech unit representation and better prosody control and further the formulation of automatic segmentation procedures for speech inventory construction and prosody modelling.
The overall project goals and research topics are: statistical models of Czech language, theory of optimal construction of dialogue in the case of man-machine communication, application of phonetic research for synthetic voice and for speech recognition by computer, parametrization and coding of speech signals, segmentation and labelling of speech signal, synthetic voice for continuous speech with prosody and high intelligibility, continuous speech recognition, voice communication and speech recognition in noisy environment, design of equipment for handicapped people.
Jan Uhlír - Czech Technical University
Tel: +42 2 24352850