• The Journal of the Acoustical Society of Korea
• /
• v.18 no.5
• /
• pp.21-27
• /
• 1999

In this paper, we propose an improved MLP-based English pronunciation dictionary generator to apply to the variable vocabulary word recognizer. The variable vocabulary word recognizer can process any words specified in Korean word lexicon dynamically determined according to the current recognition task. To extend the ability of the system to task for English words, it is necessary to build a pronunciation dictionary generator to be able to process words not included in a predefined lexicon, such as proper nouns. In order to build the English pronunciation dictionary generator, we use context-dependent grapheme-to-phoneme multi-layer perceptron(MLP) architecture for each grapheme. To train each MLP, it is necessary to obtain grapheme-to-phoneme training data from general pronunciation dictionary. To automate the process, we use dynamic programming(DP) algorithm with some distance metrics. For training and testing the grapheme-to-phoneme MLPs, we use general English pronunciation dictionary with about 110 thousand words. With 26 MLPs each having 30 to 50 hidden nodes and the exception grapheme lexicon, we obtained the word accuracy of 72.8% for the 110 thousand words superior to rule-based method showing the word accuracy of 24.0%.

• Journal of Korea Multimedia Society
• /
• v.8 no.2
• /
• pp.181-191
• /
• 2005

The classification error of 'ㅁ', 'ㅇ' is one of the main causes of incorrect recognition in Korean characters, but there haven't been enough researches to solve this problem. In this paper, a new feature extraction method from Korean grapheme is proposed to recognize 'ㅁ', 'ㅇ'effectively. First, we defined an optimal neighbor-distance selection measure using modified Hausdorff distance, which we determined the optimal neighbor-distance by. And we extracted neighbor-angle feature which was used as the effective feature to classify the two graphemes 'ㅁ', 'ㅇ'. Experimental results show that the proposed feature extraction method worked efficiently with the small number of features and could recognize the untrained patterns better than the conventional methods. It proves that the proposed method has a generality and stability for pattern recognition.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.7
• /
• pp.656-664
• /
• 2002

We present a statistical analysis of Korean phonological variations using a Grapheme-to-Phoneme (GPT) system. The GTP system used for experiments generates pronunciation variants by applying rules modeling obligatory and optional phonemic changes and allophonic changes. These rules are derived form morphophonological analysis and government standard pronunciation rules. The GTP system is optimized for continuous speech recognition by generating phonetic transcriptions for training and constructing a pronunciation dictionary for recognition. In this paper, we describe Korean phonological variations by analyzing the statistics of phonemic change rule applications for the 60,000 sentences in the Samsung PBS Speech DB. Our results show that the most frequently happening obligatory phonemic variations are in the order of liaison, tensification, aspirationalization, and nasalization of obstruent, and that the most frequently happening optional phonemic variations are in the order of initial consonant h-deletion, insertion of final consonant with the same place of articulation as the next consonants, and deletion of final consonant with the same place of articulation as the next consonant's, These statistics can be used for improving the performance of speech recognition systems.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.530-536
• /
• 2021

Recently, due to the development of deep learning, end-to-end speech recognition, which directly maps graphemes to speech signals, shows good performance. Especially, among the end-to-end models, conformer shows the best performance. However end-to-end models only focuses on the probability of which grapheme will appear at the time. The decoding process uses a greedy search or beam search. This decoding method is easily affected by the final probability output by the model. In addition, the end-to-end models cannot use external pronunciation and language information due to structual problem. Therefore, in this paper conformer with lexicon transducer is proposed. We compare phoneme-based model with lexicon transducer and grapheme-based model with beam search. Test set is consist of words that do not appear in training data. The grapheme-based conformer with beam search shows 3.8 % of CER. The phoneme-based conformer with lexicon transducer shows 3.4 % of CER.

• Journal of the Korea Society of Computer and Information
• /
• v.23 no.1
• /
• pp.75-84
• /
• 2018

In this study, the researchers attempted to automate the process of painting the characters on the road surface, which is currently done by manual labor, by using the information and communication technology. Here are the descriptions of how we put in our efforts to achieve such a goal. First, we familiarized ourselves with the current regulations about painting letters or characters on the road, with reference to Road Mark Installation Management Manual of the National Police Agency. Regarding the graphemes, we adopted a new one using connection components, in Gothic print characters which was within the range of acceptance according to the aforementioned manual. We also made it possible for the automated program to recognize the graphemes by means of the feature dots of the isolated dots, end dots, 2-line gathering dots, and gathering dots of 3 lines or more. Regarding the database, we built graphemes database for plotting information, classified the characters by means of the arrangement information of the graphemes and the layers that the graphemes form within the characters, and last but not least, made the character shape information database for character plotting by using such data. We measured the layers and the arrangement information of the graphemes consisting the characters by using the information of: 1) the information of the position of the center of gravity, and 2) the information of the graphemes that was acquired through vertical exploration from the center of gravity in each grapheme. We identified and compared the group to which each character of the database belonged, and recognized the characters through the use of the information gathered using this method. We analyzed the input characters using the aforementioned analysis method and database, and then converted into plotting information. It was shown that the plotting was performed after the correction.

• Journal of Information Processing Systems
• /
• v.8 no.4
• /
• pp.653-668
• /
• 2012

In this study, an improved HMM based recognition model is proposed for online English and Korean handwritten characters. The pattern elements of the handwriting model are sub character strokes and ligatures. To deal with the problem of handwriting style variations, a modified Hierarchical Clustering approach is introduced to partition different writing styles into several classes. For each of the English letters and each primitive grapheme in Korean characters, one HMM that models the temporal and spatial variability of the handwriting is constructed based on each class. Then the HMMs of Korean graphemes are concatenated to form the Korean character models. The recognition of handwritten characters is implemented by a modified level building algorithm, which incorporates the Korean character combination rules within the efficient network search procedure. Due to the limitation of the HMM based method, a post-processing procedure that takes the global and structural features into account is proposed. Experiments showed that the proposed recognition system achieved a high writer independent recognition rate on unconstrained samples of both English and Korean characters. The comparison with other schemes of HMM-based recognition was also performed to evaluate the system.

• Proceedings of the Korean Information Science Society Conference
• /
• 2002.04b
• /
• pp.646-648
• /
• 2002

본 논문에서는 차량번호판의 용도구분 문자를 자소 단위로 분리하는 효율적인 방법을 제안하고, 신경망을 이용하여 자소를 인식하는 방법을 소개한다. 용도구분 문자(가, 거, 나, 너‥‥)는 실제 번호판의 훼손, 카메라의 성능, 기타 여러 가지 조건에 의해서 번호판 영상에 많은 잡영이 포함된다. 따라서 차량번호판 한글문자를 자소분리하는 것은 어려운 작업이다. 제안하는 이진 영상처리 기법(morphological operation, connected component labeling 등) 으로 분리된 자소가 인식시스템으로의 입력벡터로 입력되었을 때 높은 인식률을 보이는 것을 실험을 통하여 확인하였다

• Proceedings of the KIEE Conference
• /
• 2009.07a
• /
• pp.1919_1920
• /
• 2009

한글 문자 인식에 관한 연구는 통계적 방법과 구조적 방법, 신경 회로망 등 다양한 방법론이 제시되어 왔다. 그러나 한글은 영문이나 숫자에 비해 방대한 문자수와 복잡한 구조로 인하여 인식에 많은 어려움을 가지고 있다. 따라서 본 논문에서는 한글을 가장 단순한 구조인 자음과 모음으로 분리한 뒤 각 개체의 픽셀 분포를 파악하고, 한글의 구조적 특징을 이용하여 자소의 행과 열에서의 peak값과 픽셀의 분포를 그룹으로 나누어 한글을 인식하는 방법을 제시한다.

• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.49-55
• /
• 2018

This paper aims to describe a knowledge-based pronunciation generation system for French. It has been reported that a rule-based pronunciation generation system outperforms most of the data-driven ones for French; however, only a few related studies are available due to existing language barriers. We provide basic information about the French language from the point of view of the relationship between orthography and pronunciation, and then describe our knowledge-based pronunciation generation system, which consists of morphological analysis, Part-of-Speech (POS) tagging, grapheme-to-phoneme generation, and phone-to-phone generation. The evaluation results show that the word error rate of POS tagging, based on a sample of 1,000 sentences, is 10.70% and that of phoneme generation, using 130,883 entries, is 2.70%. This study is expected to contribute to the development and evaluation of speech synthesis or speech recognition systems for French.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.04c
• /
• pp.244-246
• /
• 2003

본 논문에서는 낱자 인식기와 자소 조합 인식기를 혼용한 저품질 인쇄체 한글의 고성능 인식 방법을 제안하였다. 제안한 방법에서는 입력 문자를 한글 6형식과 기타 형식의 문자, 총 7종으로 분류한, 입력문자를 인식 대상 문자의 수와 자소 복잡도에 따라 하나 또는 두 개의 인식 단위(HRU: Hangul recognition unit)로 분리하여 인식한다. 각 인식 단위 영상에서 추출한 방향각 특징을 다층신경망 인식기를 이용하여 인식한다. 다음으로, 각 다층신경망 인식기의 신뢰도를 조합하여 최종 인식 결과를 도출한다. 제안한 방법을 사용한 실험에서 98.80%의 인식률을 얻을 수 있었으며, 이는 기존 방법에 비해 23.61%의 오류가 감소한 것이다.