Consonantal Recognition Using SVM and New Hierarchical Decision Structure Based in the Articulatory Phonetics

In this work, a new concept of making the consonantal recognition is proposed. We used several units (phonemes, diphones and syllables) to make the word recognition. This concept was carried out by a new hierarchical decision structure, based on the articulatory phonetics and SVM (support vector machine). The speech features used were MFCC (mel-frequency cepstral coefficient) and WPT (waveletpacket transform). Eighteen consonantal phonemes have been used in the recognition. The database used for the recognition was a set of two-syllable words of the Brazilian Portuguese language. The experimental results showed success rates of 98.41% for the user dependent case.