• Journal of KIISE:Software and Applications
• /
• v.26 no.1
• /
• pp.178-178
• /
• 1999

Most stochastic taggers can not resolve some morphological ambiguities that can be resolved only by referring to lexical contexts because they use only contextual probabilities based ontag n-grams and lexical probabilities. Existing lexical rules are effective for resolving such ambiguitiesbecause they can refer to lexical contexts. However, they have two limitations. One is that humanexperts tend to make erroneous rules because they are deterministic rules. Another is that it is hardand time-consuming to acquire rules because they should be manually acquired. In this paper, wepropose context-dependent lexical rules, which are lexical rules based on the statistics of a taggedcorpus, and an ambiguity-driven teaming method, which is the method of automatically acquiring theproposed rules from a tagged corpus. By using the proposed rules, the proposed tagger can partiallyannotate an unseen corpus with high accuracy because it is a kind of memorizing tagger that canannotate a training corpus with 100% accuracy. So, the proposed tagger is useful to improve theaccuracy of a stochastic tagger. And also, it is effectively used for detecting and correcting taggingerrors in a manually tagged corpus. Moreover, the experimental results show that the proposed methodis also effective for English part-of-speech tagging.

• Journal of Digital Contents Society
• /
• v.6 no.1
• /
• pp.89-94
• /
• 2005

In machine translation the goal of natural language generation is to produce an target sentence transmitting the meaning of source sentence by using an parsing tree of source sentence and target expressions. It provides generator with linguistic structures, word mapping, part-of-speech, lexical information. The purpose of this study is to research the Korean Characteristics which could be used for the establishment of an algorism in speech recognition and composite sound. This is a part of realization for the plan of automatic machine translation. The stage of MT is divided into the level of morphemic, semantic analysis and syntactic construction.

• Phonetics and Speech Sciences
• /
• v.4 no.3
• /
• pp.13-23
• /
• 2012

The sound /w/ has been traditionally regarded as an independent segment in Korean regardless of the phonological contexts in which it occurs. There have been, however, some questions regarding whether it is an independent phoneme in /CwV/ context (cf. Kang 2006). The present pilot study examined how Korean /w/ is realized in $/S^*wV/$ context by performing some perception tests. Our assumption was that if Korean /w/ is a part of the preceding complex consonant like $/C^w/$, it should be more or less uniformly articulated and perceived as such. If /w/ is an independent segment, it will be realized with speaker variability. Experiments I and II examined the identification rates as "labialized" of the spliced original stimuli of $/S^*-V/$ and $/S^{w*}-^wV/$, and the cross-spliced stimuli $/S^{w*}-V/$ and $/S^*-^wV/$. The results showed that round qualities of /w/ are perceived at significantly different temporal point with speaker and context variability. We therefore conclude that /w/ in $/S^*wV/$ context is an independent segment, not a part of the preceding segment. Full-scale examination of the production test in the future should be performed to verify the conclusion we suggested in this paper.

• Phonetics and Speech Sciences
• /
• v.1 no.2
• /
• pp.37-42
• /
• 2009

The purpose of the present study was to determine the effect of the management program known as vocal function exercise (VFE) on voice quality. Typical VFE was modified and applied to patients with vocal nodules by controlling intensity of voice and relieving the vocal fold to solve hyperfunctional problems in VFE. Eight female subjects aged between 28 and 54 who had been diagnosed with vocal nodules took part in the study. The patients performed VFEs once a week for eight weeks. Vocal function exercises consist of voice hygiene, respiratory training, phonation training, and glide training. The subjects' voices were analyzed pre and post therapy on the aspects of acoustics, maximum phonation time (MPT), GRBAS, and voice handicap index (VHI). As a result, it was found that fundamental frequency ($F_o$) was significant increased, shimmer decreased remarkably and that noise to harmonic ratio (NHR) lowered obviously in the acoustic parameter. In addition, MPT was increased significantly. The scale of GRBAS indicated significant improvement in grade, roughness, and strained voice. VHI indicated significant improvement in an emotional part. In conclusion, VFE was effective in improving voice quality for patients with vocal nodules.

• The Journal of the Acoustical Society of Korea
• /
• v.31 no.7
• /
• pp.427-434
• /
• 2012

The overlap-add technique based on waveform similarity (WSOLA) method is known to be an efficient high-quality algorithm for time scaling of speech signal. The computational load of WSOLA is concentrated on the repeated normalized cross-correlation (NCC) calculation to evaluate the similarity between two signal waveforms. To reduce the computational complexity of WSOLA, this paper proposes a fast NCC computation method, in which NCC is obtained through pre-calculated sum tables to eliminate redundancy of repeated NCC calculations in the adjacent regions. While the denominator part of NCC has much redundancy irrespective of the time-scale factor, the numerator part of NCC has less redundancy and the amount of redundancy is dependent on both the time-scale factor and optimal shift value, thereby requiring more sophisticated algorithm for fast computation. The simulation results show that the proposed method reduces about 40%, 47% and 52% of the WSOLA execution time for the time-scale compression, 2 and 3 times time-scale expansions, respectively, while maintaining exactly the same speech quality of the conventional WSOLA.

• KIISE Transactions on Computing Practices
• /
• v.22 no.7
• /
• pp.315-319
• /
• 2016

Recurrent neural network based language models (RNN LM) have shown improved results in language model researches. The RNN LMs are limited to post processing sessions, such as the N-best rescoring step of the wFST based speech recognition. However, it has considerable vocabulary problems that require large computing powers for the LM training. In this paper, we try to find the 1st pass N-gram model using word embedding, which is the simplified deep neural network. The class based language model (LM) can be a way to approach to this issue. We have built class based vocabulary through word embedding, by combining the class LM with word N-gram LM to evaluate the performance of LMs. In addition, we propose that part-of-speech (POS) tagging based LM shows an improvement of perplexity in all types of the LM tests.

• The KIPS Transactions:PartB
• /
• v.13B no.5 s.108
• /
• pp.553-560
• /
• 2006

Speech recognition has recently been investigated to produce more reliable recognition results in a noisy environment, by integrating diverse sources of information into the result derivation-level or producing new results through post-processing the prior recognition results. In this paper we propose a method which uses the user's usage patterns and the context information in speech command recognition for personal mobile devices to improve the recognition accuracy in a noisy environment. Sequential usage (or speech) patterns prior to the current command spoken are used to adjust the base recognition results. For the context information, we use the relevance between the current function of the device in use and the spoken command. Our experiment results show that the proposed method achieves about 50% of error correction rate over the base recognition system. It demonstrates the feasibility of the proposed method.

• Journal of Sinology and China Studies
• /
• v.80
• /
• pp.177-197
• /
• 2019

The question 'YOUMEIYOU(有没有)+VP?' indicating 'Perfect' was mainly used in the Southern Chinese dialect, recently this question 'YOUMEIYOU(有没有)+VP?' has also been widely used in Mandarin. In the academic world, it is reported that the appearance of 'YOUMEIYOU(有没有)+VP?' was influenced by the Southern Chinese dialect. However, there are divergent opinions about the part of speech of 'YOU(有)', such as 'auxiliary verb' , 'adverb' , 'verb', and 'particle' etc,, and these are classifiable into two main theories-'auxiliary verb' theory and 'adverb' theory. In this paper, we considered the validity of these two theories in terms of syntactic, semantic, diachronous, and typological. The results are as follows. First, based on the prototype category theory, we reclassified typical syntactic and untypical syntactic features of 'auxiliary verb' and 'adverb'. Through this, we considered the syntactic features of 'YOU(有)' are closer to the syntactic features of either of the two. As a result, It is reasonable to think that '有(YOU)' is an 'adverb' that has untypical syntactic features (It is 'adverb', but can constitute a positive-negative question, such as 'ZAIBUZAI(再不再)', 'CHANGBUCHANG(常不常)', 'CENGBUCENG(曾不曾)' and so on.). Second, the semantic features of 'YOU(有)' in the question 'YOUMEIYOU(有没有)+VP?' indicates 'Aspect' meaning ('Perfect'), does not indicate the 'Modality' meaning represented by the general 'auxiliary verb'(ability, will, deontic, epistemic etc.). Unlike English, contemporary Mandarin does not have a separate form of 'auxiliary verb'(have, be) that represents 'Modality' and 'Aspect'. On the other hand, it is reasonable to consider 'You(有)' as an 'adverb', because 'adverb' has played a role of 'Aspect' meaning(Ceng(曾), Yi(已), Zai(在), Yao(要) etc.) from the pre-QIN(先秦) dynasty before the appearance of 'Perfect particle' '了'. Third, in the Chinese history and grammar academic world, the prevailing view is that 'MEIYOU(没有)' which began to appear in front of VP in Ming(明) dynasty is regarded as a 'negative adverb'. Therefore, if 'MEIYOU(没有)' in the question 'YOUMEIYOU(有没有)+VP?' is regarded as a 'negative adverb', it is not reasonable to regard the remaining '有(YOU)' as an auxiliary verb. Also, there are some examples of 'You(有)' written as an 'adverb' in the pre-QIN(先秦) dynasty, so it is reasonable to regard '有(YOU)' as an 'adverb'. Fourth, according to the study of the linguistic typological theory, the 'H-POSSESSIVE' verb was grammaticalized toward 'Perfect' marker in many languages around the world. However, after this grammaticalization the part of speech is not all 'auxiliary verb'. In some languages, they can be grammaticalized toward 'adverb'. Therefore, it is reasonable to regard '有(YOU)' as an 'adverb'.

• Proceedings of the KSPS conference
• /
• 1996.10a
• /
• pp.421-425
• /
• 1996

A multimedia Version of ${\ulcorner}Lee-Kim test of Korean Articulation{\lrcorner}$ "Lee-Kim Test of Korean Articulation" consisting of picture test, Sentence test, user's manual and notation, analysis sheets was published in 1990 to serve as a standard tool for testing and analysing the articulation errors of normal and abnormal speakers. It has been found, however that, the picture and sentences test using the printed version of Lee-Kim test of Korean Articulation revealed several limitations, in i, e, a) inefficiency in inducing desirable response from the informants b) lack of concentration and interest on the part of informants c) no consistent way of providing the informant with a clue in case the informant is unfamiliar with the word represented by the picture or the sentence. d) no reliable means for the speech-language pathologist to analyze and evaluate the informant's speech in relation to the standard pronunciation A multimedia version of Lee-Kim Korean articulation Test which features picture and word as well as recorded voice has been developed with a view to eliminating the limitation mentioned above and facilitating the articulation test-with ease and accuracy.

• The KIPS Transactions:PartB
• /
• v.12B no.4 s.100
• /
• pp.437-442
• /
• 2005

For the purpose of speaker normalization in speaker independent speech recognition systems, experiments are conducted on a method based on Gaussian mixture model(GMM). The method, which is an improvement of the previous study based on vector quantizer, consists of modeling the probability distribution of canonical feature vectors by a GMM with an appropriate number of clusters, and of estimating the warp factor of a test speaker by making use of the obtained probabilistic model. The purpose of this study is twofold: improving the existing ML based methods, and comparing the performance of what is called 'soft decision' method with that of the previous study based on vector quantizer. The effectiveness of the proposed method is investigated by recognition experiments on the TIMIT corpus. The experimental results showed that a little improvement could be obtained tv adjusting the number of clusters in GMM appropriately.