• MALSORI
• /
• no.spc1
• /
• pp.154-166
• /
• 2002

• The Journal of the Acoustical Society of Korea
• /
• v.23 no.2
• /
• pp.173-179
• /
• 2004

Due to the insufficiency of training data in large vocabulary continuous speech recognition, similar context dependent phones can be clustered by decision trees to share the data. When the decision trees are built and used to predict unseen triphones, a phonetic question set is required. The phonetic question set, which contains categories of the phones with similar co-articulation effects, is usually generated by phonetic or linguistic experts. This knowledge-based approach for generating phonetic question set, however, may reduce the homogeneity of the clusters. Moreover, the experts must adjust the question sets whenever the language or the PLU (phone-like unit) of a recognition system is changed. Therefore, we propose a data-driven method to automatically generate phonetic question set. Since the proposed method generates the phone categories using speech data distribution, it is not dependent on the language or the PLU, and may enhance the homogeneity of the clusters. In large vocabulary speech recognition experiments, the proposed algorithm has been found to reduce the error rate by 14.3%.

• MALSORI
• /
• no.19_20
• /
• pp.43-57
• /
• 1990

The teaching of pronunciation is one of the areas in which linguistic phonetics can play an extremely useful role. This paper is concerned with the application of the results of my phonetic research to the actual teaching of Korean standard pronunciation with special reference to speech rhythm. It has been found that Korean words and utterances of various lengths are pronounced in standard Korean with one of the four main rhythmic patterns, each containing a strong stress. Unless we get the rhythmic patterns right in pronouncing Korean words and utterances, therefore, the resulting pronounciation is bound to sound dialectal or incorrect and in many instances even unintelligible to listeners. Hence the undeniable need to devise a useful technique to teach the Korean speech rhythm in a systematic way. In this paper each of the four main rhythmic patterns is presented and elaborated with sample examples taken from the living Korean. It is hoped that these examples of words and utterances can be used at the same time as useful pronunciation drill material not only for Koreans with dialectal background but also for foreign learners of Korean.

• Proceedings of the KSPS conference
• /
• 1996.10a
• /
• pp.117-122
• /
• 1996

This paper proposes a derivational account of tensing and neutralization of obstruents in Korean within the theory of Government Phonology (GP) (Kaye, Lowenstamm and Vergnaud 1990, henceforth KLV; Park 1996). We begin by outling the relevant tensing and neutralization data in Korean. We point out several problems that need to be addressed in any account of these data. We then set out the central notions of GP, pointing out how adherence to the requirement that government relations remain constant throughout a derivation under the Projection Principle prevents a GP account of tensing and neutralization in Korean, which requires government relations to switch between lexical and phonetic representations. To address this problem, we propose abandoning the Projection Principle, extending lexical representations in GP along the lines of the Markedness Theory approach (Michaels 1989), and adopting the economy principles for derivation of the Minimalist approach (Chomsky 1993; Chomsky & Lasnik 1991). finally, we summarize the analysis of obstruent phenomena in Korean within GP extended in these ways.

• MALSORI
• /
• no.21_24
• /
• pp.144-175
• /
• 1992

This paper aims to discuss determining factors of Korean allophones and to formulate and classify Korean allophonic rules systematically. The relationship between allophones and coarticulation, the most. influential factor of allophonic variation, is thoroughly investigated. Other factors -- speech tempo and style, dialect, and social factors such as age, set, class etc. -- are also briefly discussed. Allophonic rules are classified into two groups -- 3) those relevant to coarticulation and 2) those irrelevant to coarticulation. Rules of the first group are further classified into four subgroups according to the directionality of the coarticulation. Each allophonic nile formulation is explained and discussed in detai1. The allophonic rules formulated and classified in this paper are 1) Devoicing of Voiced Consonants, 2) Devoicing of Vowels, 3) Nasal Approach and Lateral Approach, 4) Uvularization, 5) Palatalization, 6) Voicing of Voiceless Lax Consonants, 7) Frication, 8) Labialization, 9) Nasalization, 10) Release Withholding and Release Masking, 11) Glottalization, 12) Flap Rule, 13) Vowel Weakening, and 14) Allophones of /ㅚ, ㅟ, ㅢ/ (which are realized as diphthongs or as monophthongs depending on phonetic contexts).

• Proceedings of the Korean Society for Language and Information Conference
• /
• 1998.02a
• /
• pp.283-294
• /
• 1998

• Proceedings of the KSPS conference
• /
• 1996.02a
• /
• pp.157-172
• /
• 1996

This study aims to suggest some theoretical methodology for analysis of the prosodic patterns in Korean Read Speech. The engineering effort relevant to the phonetic study has focused to the importance of prosodic phrasing which may play a major role in analyzing the phonetic DB. Before establishing the prosodic phrase as the prosodic unit, we should describe the features of the boundary signal in a target sentence. With this in mind, the general characteristics of Read Speech and the ToBI(tones and Break Indices), which has been currently in vogue with respect to the prosodic labelling system were presented as the first step. The concrete analysis was carried out with the fable 'North Wind and the Sun' Korean version, where about 25 prosodic units were discriminated by perceptual approach for 5 subjects. Establishing various informations which can be used for deciding a boundary position systematically, we can proceed to the next, viz. acoustic analysis of prosodic unit. The most important which we primarily study for improving the naturalness of synthetic speech may be, at first, detecting the boundary signals in the speech file and accordingly reestablishment it within the raw text.

• Journal of Computing Science and Engineering
• /
• v.5 no.3
• /
• pp.223-235
• /
• 2011

Performing approximate data matching has always been an intriguing problem for both industry and academia. This task becomes even more challenging when the requirement of data privacy rises. In this paper, we propose a novel technique to address the problem of efficient privacy-preserving approximate record linkage. The secure framework we propose consists of two basic components. First, we utilize a secure blocking component based on phonetic algorithms statistically enhanced to improve security. Second, we use a secure matching component where actual approximate matching is performed using a novel private approach of the Levenshtein Distance algorithm. Our goal is to combine the speed of private blocking with the increased accuracy of approximate secure matching.

• MALSORI
• /
• no.44
• /
• pp.115-130
• /
• 2002

Voice input is often required in many new application environments, but the low rate of speech recognition makes it difficult to extend its application. Previous approaches were to raise the accuracy of the recognition by post-processing of the recognition results, which were all lexical-oriented. We suggest a new semantic-oriented approach in speech recognition error correction. Through experiments using a speech-driven in-vehicle telematics information application, we show the excellent performance of our approach and some advantages it has as a semantic-oriented approach over a pure lexical-oriented approach.

• Proceedings of the KSPS conference
• /
• 2005.11a
• /
• 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.