Speech pattern recognition apparatus utilizing multiple independent sequences of phonetic segments

An A/D converter quantizes an input speech signal, a sound analyzing section provides an LPC sound analysis to quantized data, and feature parameters are extracted. A matching section obtains PS sequences in the 1st through nth places (n: an integer which is larger than 1) from the obtained feature parameters by use of a PS dictionary. An HMM learning section parameter-estimation learns the obtained PS sequences in the 1st through nth places by Baum Welch algorithm of a hidden Markov model as an individual learning PS sequence.