Algorithm for Developing Speech Components for Educational Software with Acoustic Feedback

The paper presents the algorithm of developing speech components aimed to be implemented into educational systems with elements of acoustic feedback. Such systems are demanded in the domain of distant education as far as foreign language teaching and pronunciation training is concerned. The need of such systems is evident because human teacher is not always at hand when the student needs phonetic training. Besides, natural pronunciation is not always achieved by imitation exercises which could be done by imitation authentic speech recorded on the tapes, CDs and video. But nowadays teacher’s feedback is required to correct student’s pronunciation in case there are errors. The aim of the project conducted in Moscow State Linguistic university is to substitute human teacher by a computer software to some extent in distant education environment. Thus the algorithms of speech recognition detecting pronunciation deviation are analyzed. Moreover, the presented paper contains an overview of the most frequent problems which occur while recognizing interfered speech (in case of RussianEnglish interference). While working on the project a database of the most frequent recognition errors in recognizing English speech of native Russian is being built. The data will definitely turn out handy in further work on the project. Some preliminary results and methodology are also presented in the paper.