Using Fuzzy Logic and Features Measured from the Time Domain to Achieve Smart Separation of Phonetic Units

The segmentation of uttered speech into phonetic units is a key processing task for successfully implementing speech recognition systems. This paper presents a smart approach to phonetic segmentation of uttered speech that separates vowels from consonants. Time-domain feature-extraction algorithms are applied to speech to extract features at minimum computational cost. Fuzzy decision logic is used to infer the effective separation point, considering coarticulations specific to uttered speech. Experimental results have shown this approach to be effective in separating phonetic units, while requiring minimal computing power and reducing system complexity. Key-Words: Speech analysis, speech segmentation, speech recognition, fuzzy decision logic, zero-crossing rate, speech energy, pitch estimation.