Speech recognition through phoneme segmentation and neural classification

The problem of speech recognition may be viewed as the identification of the basic components (phonemes) of the human speech through a high-level measurement procedure working on segments of the vocal signal, their classification, and the identification of their combination into the individual words so as to derive the complete word identification. The efficient and effective solution of the whole problem has several applications (e.g., the automatic typewriter) as well as the simple phoneme recognition may be exploited for some innovative areas (e.g., the voice compression for telecommunication). This paper shows the use of hybrid soft-computing techniques for the signal segmentation and the phoneme classification and the application to voice compression.