Aplikasi Pengenalan UcapanBerdasarkan Suku Kata Konsonan-VokalMenggunakan Algoritma Hidden Markov Model

Abstract – There are a lot of speech recognition methods; the simplest one is by recognize every single word. But, this method has a weakness which is it needs a big memory to be used with various words. This can happen because of numbers of words which can be recognized are the same as numbers of words which are used to get recognition parameters. To handle this problem, we build recognition system based on their syllables. By using this system, speech will be recognized based on their syllables so speech inputs which are used to get recognition parameters are fewer than recognition system which are used words to get recognition parameters. In this thesis we build a system to recognized speech based on their syllables with Hidden Markov Model (HMM) algorithm. First of all, words which have consonant-vocal syllables are recorded. Those words will be segmented for every syllable, so we get consonant-vocal syllables. Then those syllables trained by HMM algorithm to get recognition parameters. Next process, same as the first process—recording words which are have consonantvocal syllables, then segmenting those words into syllables. Every syllable by segmentation process will be counted their recognition probabilities with HMM algorithm. Syllable with the highest probability score is the recognized syllable. Output of this system is text based on speech recognition. After do some tests for this program, the results are the highest speech recognition percentage for speech recognition is happen when recognize trained speech and the score is 85,25 %. The speech recognition percentage for recognize outside trained speech is 61,65 % and the speech recognition percentage for recognize speech which is recorded by this program is 53,9 %. For recognition system which used syllable as its recognition parameters,segmentation process will influence system’s ability to recognize speech.