Best Wavelet Filter for a Wavelet Neural Fricatives Recognition System

摘要: Direct recognition of phonemes in speaker independent speech systems still cannot guarantee good enough results. But grouping at first then trying to recognize the phoneme itself is a promising field. On other hand wavelets are widely used and systems, this motivated by ability wavelet coefficients capture important time frequency features. In work effect filter type on efficiency system investigated (specifically fricatives). The Probabilistic neural network was as pattern matching stage for its well known power full solving classification problems. It found that Daubechies family (generally from db15 db23) candidate fricatives based feature extraction stage.