• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2005년도 추계 학술대회 발표논문집
• /
• pp.103-106
• /
• 2005

In this paper, we propose a language modeling approach to improve the performance of a large vocabulary continuous speech recognition system. The proposed approach is based on the active learning framework that helps to select a text corpus from a plenty amount of text data required for language modeling. The perplexity is used as a measure for the corpus selection in the active learning. From the recognition experiments on the task of continuous Korean speech, the speech recognition system employing the language model by the proposed language modeling approach reduces the word error rate by about 6.6 % with less computational complexity than that using a language model constructed with randomly selected texts.

• 한국멀티미디어학회논문지
• /
• 제4권6호
• /
• pp.524-533
• /
• 2001

연속음성인식 시스템의 실용화를 위해서 가장 중요한 것은 높은 인식 성능을 가지면서 동시에 실시간으로 인식되어야 한다. 이를 위하여 본 연구에서는 먼저 연속음성인식의 인식률 향상을 위하여 효과적인 음향모델을 구성하기 위하여 PDT-SSS(Phonetic Decision Tree-based Successive State Splitting) 알고리즘을 도입하여 HM-Net을 구성하고, 언어모델로서 반복학습을 이용하여 인식률 향상을 제고한다. 그리고, 기존의 연구에서 유효함이 입증된 프레임 단위 적응 프루닝 알고리즘을 연속음성에 적용하여 인식 속도를 개선하고자 한다. 제안된 방법의 유효성을 확인하기 위하여, 남성 4인이 항공편 예약 관련 음성에 대하여 인식 실험을 수행하였다. 그 결과 연속음성인식률 90.9%, 단어인식률 90.7%의 높은 인식성능을 얻었으며, 적응 프루닝 알고리즘을 적용한 경우 인식성능의 저하없이 약 1.2초(전체의 15%)의 인식시간을 줄일 수 있어 제안된 방법의 유효성을 확인할 수 있었다.

• 음성과학
• /
• 제10권1호
• /
• pp.71-84
• /
• 2003

In the acoustic modeling for large vocabulary speech recognition, a sparse data problem caused by a huge number of context-dependent (CD) models usually leads the estimated models to being unreliable. In this paper, we develop a new clustering method based on the C45 decision-tree learning algorithm that effectively encapsulates the CD modeling. The proposed scheme essentially constructs a supervised decision rule and applies over the pre-clustered triphones using the C45 algorithm, which is known to effectively search through the attributes of the training instances and extract the attribute that best separates the given examples. In particular, the data driven method is used as a clustering algorithm while its result is used as the learning target of the C45 algorithm. This scheme has been shown to be effective particularly over the database of low unknown-context ratio in terms of recognition performance. For speaker-independent, task-independent continuous speech recognition task, the proposed method reduced the percent accuracy WER by 3.93% compared to the existing rule-based methods.

• 대한음성학회지:말소리
• /
• 제59호
• /
• pp.53-68
• /
• 2006

We report the evaluation results of the Korean speech recognition platform called ECHOS. The platform has an object-oriented and reusable architecture so that researchers can easily evaluate their own algorithms. The platform has all intrinsic modules to build a large vocabulary speech recognizer: Noise reduction, end-point detection, feature extraction, hidden Markov model (HMM)-based acoustic modeling, cross-word modeling, n-gram language modeling, n-best search, word graph generation, and Korean-specific language processing. The platform supports both lexical search trees and finite-state networks. It performs word-dependent n-best search with bigram in the forward search stage, and rescores the lattice with trigram in the backward stage. In an 8000-word continuous speech recognition task, the platform with a lexical tree increases 40% of word errors but decreases 50% of recognition time compared to the HTK platform with flat lexicon. ECHOS reduces 40% of recognition errors through incorporation of cross-word modeling. With the number of Gaussian mixtures increasing to 16, it yields word accuracy comparable to the previous lexical tree-based platform, Julius.

• The Journal of the Acoustical Society of Korea
• /
• 제15권3E호
• /
• pp.60-65
• /
• 1996

This paper describes on the Korean continuous speech recognition system for flight reservation. It adopts a frame-synchronous One-Pass DP search algorithm driven by syntactic constraints of context free grammar(CFG). For recognition, 48 phoneme-like units(PLU) were defined and used as basic units for acoustic modeling of Korean. This modeling was conducted using a HMM technique, where each model has 4-states 3-continuous output probability distributions and 3-discrete-duration distributions. Language modeling by CFG was also applied to the task domain of flight reservation, which consisted of 346 words and 422 rewriting rules. In the tests, the sentence recognition rate of 62.6% was obtained after speaker adaptation.

• 한국음향학회지
• /
• 제21권6호
• /
• pp.543-551
• /
• 2002

이 논문에서는 20,000 단어급의 대어휘를 대상으로 실시간 연속음성 인식을 수행할 수 있는 탐색 방법을 제안한다. 기본적인 탐색 방법은 토큰 전파 방식의 비터비 (Viterbi) 디코딩 알고리듬을 이용한 1 패스로 구성된다. 언어 모델 네트워크를 도입하여 다양한 언어 모델들을 일관된 탐색 공간으로 구성하도록 하였으며, 프루닝(pruning) 단계에서 살아남은 토큰들로부터 동적으로 탐색 공간을 재구성하였다. 용이한 후처리를 위해 워드그래프 및 N개의 최적 문장을 출력할 수 있도록 비터비 알고리듬을 수정하였다. 이렇게 구성된 디코더는 20,000 단어급 데이터 베이스에 대해 테스트하였으며 인식률 및 RTF측면에서 평가되었다.

• 말소리와 음성과학
• /
• 제7권4호
• /
• pp.41-47
• /
• 2015

This paper examines whether the Global Vector model is applicable to Korean data as a universal learning algorithm. The main purpose of this study is to compare the global vector model (GloVe) with the word2vec models such as a continuous bag-of-words (CBOW) model and a skip-gram (SG) model. For this purpose, we conducted an experiment by employing an evaluation corpus consisting of 70 target words and 819 pairs of Korean words for word similarities and analogies, respectively. Results of the word similarity task indicated that the Pearson correlation coefficients of 0.3133 as compared with the human judgement in GloVe, 0.2637 in CBOW and 0.2177 in SG. The word analogy task showed that the overall accuracy rate of 67% in semantic and syntactic relations was obtained in GloVe, 66% in CBOW and 57% in SG.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 1996년도 10월 학술대회지
• /
• pp.472-472
• /
• 1996

The material of database for speech recognition should include phonetic phenomena as much as possible. At the same time, such material should be phonetically compact with low redundancy[1, 2]. The phonetic phenomena in continuous speech is the key problem in speech recognition. This paper describes the processing of a set of sentences collected from the database of 1993 and 1994 "People's Daily"(Chinese newspaper) which consist of news, politics, economics, arts, sports etc.. In those sentences, both phonetic phenometla and sentence patterns are included. In continuous speech, phonemes always appear in the form of allophones which result in the co-articulary effects. The task of designing a speech database should be concerned with both intra-syllabic and inter-syllabic allophone structures. In our experiments, there are 404 syllables, 415 inter-syllabic diphones, 3050 merged inter-syllabic triphones and 2161 merged final-initial structures in read speech. Statistics on the database from "People's Daily" gives and evaluation to all of the possible phonetic structures. In this sentence set, we first consider the phonetic balances among syllables, inter-syllabic diphones, inter-syllabic triphones and semi-syllables with their junctures. The syllabic balances ensure the intra-syllabic phenomena such as phonemes, initial/final and consonant/vowel. the rest describes the inter-syllabic jucture. The 1560 sentences consist of 96% syllables without tones(the absent syllables are only used in spoken language), 100% inter-syllabic diphones, 67% inter-syllabic triphones(87% of which appears in Peoples' Daily). There are rougWy 17 kinds of sentence patterns which appear in our sentence set. By taking the transitions between syllables into account, the Chinese speech recognition systems have gotten significantly high recognition rates[3, 4]. The following figure shows the process of collecting sentences. [people's Daily Database] -> [segmentation of sentences] -> [segmentation of word group] -> [translate the text in to Pin Yin] -> [statistic phonetic phenomena & select useful paragraph] -> [modify the selected sentences by hand] -> [phonetic compact sentence set]

• The Journal of the Acoustical Society of Korea
• /
• 제17권4E호
• /
• pp.11-16
• /
• 1998

For large vocabulary speech recognition using HMMs, context-dependent subword units have been often employed. However, when context-dependent phone models are used, they result in a system which has too may parameters to train. The problem of too many parameters and too little training data is absolutely crucial in the design of a statistical speech recognizer. Furthermore, when building large vocabulary speech recognition systems, unseen triphone problem is unavoidable. In this paper, we propose the modified phonetic decision tree algorithm for the automatic prediction of unseen triphones which has advantages solving these problems through following two experiments in Japanese contexts. The baseline experimental results show that the modified tree based clustering algorithm is effective for clustering and reducing the number of states without any degradation in performance. The task experimental results show that our proposed algorithm also has the advantage of providing a automatic prediction of unseen triphones.

• 한국음향학회지
• /
• 제22권5호
• /
• pp.388-402
• /
• 2003

In this paper, we carried out the word, 4 continuous digits. continuous, and task-independent word recognition experiments to verify the effectiveness of the re-defined phoneme-likely units (PLUs) for the phonetic decision tree based HM-Net (Hidden Markov Network) context-dependent (CD) acoustic modeling in Korean appropriately. In case of the 48 PLUs, the phonemes /ㅂ/, /ㄷ/, /ㄱ/ are separated by initial sound, medial vowel, final consonant, and the consonants /ㄹ/, /ㅈ/, /ㅎ/ are also separated by initial sound, final consonant according to the position of syllable, word, and sentence, respectively. In this paper. therefore, we re-define the 39 PLUs by unifying the one phoneme in the separated initial sound, medial vowel, and final consonant of the 48 PLUs to construct the CD acoustic models effectively. Through the experimental results using the re-defined 39 PLUs, in word recognition experiments with the context-independent (CI) acoustic models, the 48 PLUs has an average of 7.06%, higher recognition accuracy than the 39 PLUs used. But in the speaker-independent word recognition experiments with the CD acoustic models, the 39 PLUs has an average of 0.61% better recognition accuracy than the 48 PLUs used. In the 4 continuous digits recognition experiments with the liaison phenomena. the 39 PLUs has also an average of 6.55% higher recognition accuracy. And then, in continuous speech recognition experiments, the 39 PLUs has an average of 15.08% better recognition accuracy than the 48 PLUs used too. Finally, though the 48, 39 PLUs have the lower recognition accuracy, the 39 PLUs has an average of 1.17% higher recognition characteristic than the 48 PLUs used in the task-independent word recognition experiments according to the unknown contextual factor. Through the above experiments, we verified the effectiveness of the re-defined 39 PLUs compared to the 48PLUs to construct the CD acoustic models in this paper.