• The Journal of the Acoustical Society of Korea
• /
• v.28 no.3
• /
• pp.239-244
• /
• 2009

This paper is to develop the Korean Phonotactic Probability Calculator (KPPC) that anticipates the phonotactic probability in Korean. KPPC calculates the positional segment frequecncy, position-specific biphone frequency and position-specific triphone frequency. And KPPC also calculates the Neighborhood Density that is the number of words that sound similar to a target word. The Phonotactic Calculator that was developed in University of Kansas can be analyzed by the computer-readable phonemic transcription. This can calculate positional frequency and position-specific biphone frequency that were derived from 20,000 dictionary words. But KPPC calculates positional frequency, positional biphone frequency, positional triphone frequency and neighborhood density. KPPC can calculate by korean alphabet or computer-readable phonemic transcription. This KPPC can anticipate high phonotactic probability, low phonotactic probability, high neighborhood density and low neighborhood density.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.73-76
• /
• 2001

본 논문에서는 음소인식기와 음소결합확률모델을 이용하여 전화음성을 대상으로 입력음성이 어느 나라 말 인지를 식별할 수 있는 언어식별시스템을 구현하였고 성능을 실험하였다. 시스템은 음소인식기로 입력음성에 대한 음소열을 인식하는 과정, 인식된 음소열을 이용하여 인식대상 언어별 음소결합확률모델을 생성하는 훈련과정, 훈련과정에서 생성된 음소결합확률모델로부터 확률 값을 계산하여 인식결과를 출력하는 식별과정으로 구성된다. 본 논문에서는 음소결합확률모델로부터 우도를 계산할 때 정보이론(Information Theory, Shannon and Weaver, 1949)을 이용하여 가중치를 적용하는 방법을 제안하였다. 시스템의 훈련 및 실험에는 OGI 11개국어 전화음성 corpus (OGI-TS)를 사용하였으며, 음소인식기는 HTK를 이용하여 구현하였고 음소인식기 훈련에는 NTIMIT 전화음성 DB를 이용하였다. 실험결과 11개국어를 대상으로 45초 길이의 음성에 대해서 평균 $74.1\%$, 10초 길이의 음성에 대해서는 평균 $57.1\%$의 인식률을 얻을 수 있었다.

• Journal of Korea Multimedia Society
• /
• v.6 no.2
• /
• pp.216-223
• /
• 2003

In this paper, we present SCCRS(Speech and Character Combined Recognition System) for speaker /writer independent. on-line multimodal interfaces. In general, it has been known that the CHMM(Continuous Hidden Markov Mode] ) is very useful method for speech recognition and on-line character recognition, respectively. In the proposed method, the same CHMM is applied to both speech and character recognition, so as to construct a combined system. For such a purpose, 115 CHMM having 3 states and 9 transitions are constructed using MLE(Maximum Likelihood Estimation) algorithm. Different features are extracted for speech and character recognition: MFCC(Mel Frequency Cepstrum Coefficient) Is used for speech in the preprocessing, while position parameter is utilized for cursive character At recognition step, the proposed SCCRS employs OPDP (One Pass Dynamic Programming), so as to be a practical combined recognition system. Experimental results show that the recognition rates for voice phoneme, voice word, cursive character grapheme, and cursive character word are 51.65%, 88.6%, 85.3%, and 85.6%, respectively, when not using any language models. It demonstrates the efficiency of the proposed system.