Speech Recognition for Disabilities People

Abstract

In the last few years, thanks to the evolution of data processing, we attended a rather important evolution of the systems of speech recognition. The developed systems depend for applications; it has systems of isolated words recognition and continuous speech recognition systems. The systems of isolated words recognition have as application the vocal order while the systems of continuous speech recognition systems have like principal application the vocal dictation. For the disabilities people the absence of the data bases and diversity of the articulator handicaps are major obstacles for the construction of reliable speech recognition systems, which explains poverty of the market in systems of speech recognition for disabilities people. The work developed consists in adapting some of the existing systems of speech recognition to the people who have articulator handicaps. For the systems of isolated words recognition, we used a dynamic approach of training which makes it possible the system progressively to adapt to the users during his use. this approach makes it possible for disabilities user to use the system without passing by a long phase of training which is generally heavy and painful for these people, it also allows the system recognition to use a training base under the same test conditions. For the continues recognition systems we used two approaches: (1) a dynamic adaptation of the phonetic models of the continues recognition systems for the disabilities people, this approach makes it possible to adapt the system of recognition to the user and can be to use for people normal speaking. (2) Use of independent language segmentation (ALISP) for the recognition. This approach consists to use the correspondence between ALISP segmentation and the phonetics and the models of segments ALISP to build the system of recognition. For the word recognition system we used the dynamic training with DTW