• MALSORI
• /
• no.53
• /
• pp.75-92
• /
• 2005

Prosody Generation Based on C-ToBI Representation for Text-to-SpeechSeungwon Kim, Yu Zheng, Gary Geunbae Lee, Byeongchang KimProsody modeling is critical in developing text-to-speech (TTS) systems where speech synthesis is used to automatically generate natural speech. In this paper, we present a prosody generation architecture based on Chinese Tone and Break Index (C-ToBI) representation. ToBI is a multi-tier representation system based on linguistic knowledge to transcribe events in an utterance. The TTS system which adopts ToBI as an intermediate representation is known to exhibit higher flexibility, modularity and domain/task portability compared with the direct prosody generation TTS systems. However, the cost of corpus preparation is very expensive for practical-level performance because the ToBI labeled corpus has been manually constructed by many prosody experts and normally requires a large amount of data for accurate statistical prosody modeling. This paper proposes a new method which transcribes the C-ToBI labels automatically in Chinese speech. We model Chinese prosody generation as a classification problem and apply conditional Maximum Entropy (ME) classification to this problem. We empirically verify the usefulness of various natural language and phonology features to make well-integrated features for ME framework.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.488-495
• /
• 2021

We propose a speech recognition system based on conformer. Conformer is known to be convolution-augmented transformer, which combines transfer model for capturing global information with Convolution Neural Network (CNN) for exploiting local feature effectively. The baseline system is developed to be a transfer-based speech recognition using Long Short-Term Memory (LSTM)-based language model. The proposed system is a system which uses conformer instead of transformer with transformer-based language model. When Electronics and Telecommunications Research Institute (ETRI) speech corpus in AI-Hub is used for our evaluation, the proposed system yields 5.7 % of Character Error Rate (CER) while the baseline system results in 11.8 % of CER. Even though speech corpus is extended into other domain of AI-hub such as NHNdiguest speech corpus, the proposed system makes a robust performance for two domains. Throughout those experiments, we can prove a validation of the proposed system.

• Korean Journal of Cognitive Science
• /
• v.19 no.3
• /
• pp.257-281
• /
• 2008

The expression 'kes' is one of the most widely used ones in the language whose uses are highly dependent upon the context. These highly-context dependent uses make it hard to determine its grammatical properties. As a way of examining the properties in a rather controlled context, this paper collects a series of speeches made by government officials and examines the grammatical properties of the expression in the corpus. In particular, the paper, based on the 539 instances of 'kes' uses extracted from the corpus, focuses on the 7 types of 'kes' constructions most widely used in the collected speech corpus.

• Phonetics and Speech Sciences
• /
• v.8 no.3
• /
• pp.1-9
• /
• 2016

This paper investigated the phoneme distribution and phonological processes of orthographic and pronounced phrasal words in light of syllable structure in the Seoul Corpus in order to provide linguists and phoneticians with a clearer understanding of the Korean language system. To achieve the goal, the phrasal words were extracted from the transcribed label scripts of the Seoul Corpus using Praat. Following this, the onsets, peaks, codas and syllable types of the phrasal words were analyzed using an R script. Results revealed that k0 was most frequently used as an onset in both orthographic and pronounced phrasal words. Also, aa was the most favored vowel in the Korean syllable peak with fewer phonological processes in its pronounced form. The total proportion of all diphthongs according to the frequency of the peaks in the orthographic phrasal words was 8.8%, which was almost double those found in the pronounced phrasal words. For the codas, nn accounted for 34.4% of the total pronounced phrasal words and was the varied form. From syllable type classification of the Corpus, CV appeared to be the most frequent type followed by CVC, V, and VC from the orthographic forms. Overall, the onsets were more prevalent in the pronunciation more than the codas. From the results, this paper concluded that an analysis of phoneme distribution and phonological processes in light of syllable structure can contribute greatly to the understanding of the phonology of spoken Korean.

• 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.

• ETRI Journal
• /
• v.23 no.4
• /
• pp.168-176
• /
• 2001

This paper discusses two important issues of corpus-based synthesis: synthesis unit generation based on phrase break strength information and pruning redundant synthesis unit instances. First, the new sentence set for recording was designed to make an efficient synthesis database, reflecting the characteristics of the Korean language. To obtain prosodic context sensitive units, we graded major prosodic phrases into 5 distinctive levels according to pause length and then discriminated intra-word triphones using the levels. Using the synthesis unit with phrase break strength information, synthetic speech was generated and evaluated subjectively. Second, a new pruning method based on weighted vector quantization (WVQ) was proposed to eliminate redundant synthesis unit instances from the synthesis database. WVQ takes the relative importance of each instance into account when clustering similar instances using vector quantization (VQ) technique. The proposed method was compared with two conventional pruning methods through objective and subjective evaluations of synthetic speech quality: one to simply limit the maximum number of instances, and the other based on normal VQ-based clustering. For the same reduction rate of instance number, the proposed method showed the best performance. The synthetic speech with reduction rate 45% had almost no perceptible degradation as compared to the synthetic speech without instance reduction.

• Journal of Korean language education
• /
• v.29 no.4
• /
• pp.137-168
• /
• 2018

The purpose of this study is to prepare the basis for a more objective evaluation of oral fluency by comparing speech patterns of Korean native speakers and L2 Korean learners. For this purpose, the current study focused on the analysis of speech materials of the 21st century Sejong spoken corpus and Korean learner corpus. We compared the oral fluency of Korean native speakers and Korean learners based on speech rate, pause, and discourse markers. The results show that the pattern of Korean learners is different to that of Korean native speakers in all aspects of speech rate, pause, and discourse markers; even though proficiency of Korean leaners show increase, they could not reach the oral fluency level of Korean native speakers. At last, based on these results of the analysis, we added suggestions for setting the evaluation criteria of oral fluency of Korean learners.

• Phonetics and Speech Sciences
• /
• v.10 no.4
• /
• pp.111-119
• /
• 2018

This paper proposes a method to improve speech recognition performance by extracting various pronunciations of the pseudo-morpheme unit from an eojeol unit corpus and generating a new recognition unit considering pronunciation variations. In the proposed method, we first align the pronunciation of the eojeol units and the pseudo-morpheme units, and then expand the pronunciation dictionary by extracting the new pronunciations of the pseudo-morpheme units at the pronunciation of the eojeol units. Then, we propose a new recognition unit that relies on pronunciation by tagging the obtained phoneme symbols according to the pseudo-morpheme units. The proposed units and their extended pronunciations are incorporated into the lexicon and language model of the speech recognizer. Experiments for performance evaluation are performed using the Korean speech recognizer with a trigram language model obtained by a 100 million pseudo-morpheme corpus and an acoustic model trained by a multi-genre broadcast speech data of 445 hours. The proposed method is shown to reduce the word error rate relatively by 13.8% in the news-genre evaluation data and by 4.5% in the total evaluation data.

• Phonetics and Speech Sciences
• /
• v.14 no.2
• /
• pp.39-54
• /
• 2022

This study aims to clarify the prosodic differences in speech types by examining the Korean read speech and spontaneous speech in the Korean part of the L2 Korean Speech Corpus (speech corpus for Korean as a foreign language). To this end, the articulation length, articulation speed, pause length and frequency, and the average fundamental frequency values of sentences were set as variables and analyzed via statistical methodologies (t-test, correlation analysis, and regression analysis). The results found that read speech and spontaneous speech were structurally different in the form of prosodic phrases constituting each sentence and that the prosodic elements differentiating each speech type were articulation length, pause length, and pause frequency. The statistical results show that the correlation between articulation speed and articulation length was highest in read speech, explaining that the longer a given sentence is, the faster the speaker speaks. In spontaneous speech, however, the relationship between the articulation length and the pause frequency in a sentence was high. Overall, spontaneous speech produces more pauses because short intonation phrases are continuously built to make a sentence, and as a result, the sentence gets lengthened.

• Phonetics and Speech Sciences
• /
• v.7 no.4
• /
• pp.3-8
• /
• 2015

In this paper, we investigate an input dimension reduction method using continuous word vector in deep neural network language model. In the proposed method, continuous word vectors were generated by using Google's Word2Vec from a large training corpus to satisfy distributional hypothesis. 1-of-${\left|V\right|}$ coding discrete word vectors were replaced with their corresponding continuous word vectors. In our implementation, the input dimension was successfully reduced from 20,000 to 600 when a tri-gram language model is used with a vocabulary of 20,000 words. The total amount of time in training was reduced from 30 days to 14 days for Wall Street Journal training corpus (corpus length: 37M words).