• Phonetics and Speech Sciences
• /
• v.6 no.3
• /
• pp.155-164
• /
• 2014

This paper proposes a new method to determine the recognition units for large vocabulary continuous speech recognition (LVCSR) in Korean by applying unsupervised segmentation and merging. In the proposed method, a text sentence is segmented into morphemes and position information is added to morphemes. Then submorpheme units are obtained by splitting the morpheme units through the maximization of posterior probability terms. The posterior probability terms are computed from the morpheme frequency distribution, the morpheme length distribution, and the morpheme frequency-of-frequency distribution. Finally, the recognition units are obtained by sequentially merging the submorpheme pair with the highest frequency. Computer experiments are conducted using a Korean LVCSR with a 100k word vocabulary and a trigram language model obtained by a 300 million eojeol (word phrase) corpus. The proposed method is shown to reduce the out-of-vocabulary rate to 1.8% and reduce the syllable error rate relatively by 14.0%.

• The Journal of the Acoustical Society of Korea
• /
• v.20 no.2E
• /
• pp.3-8
• /
• 2001

In this paper, we will examine a Korean large vocabulary continuous speech recognition (LVCSR) system for broadcast news speech. The combined vowel and implosive unit is included in a phone set together with other short phone units in order to obtain a longer unit acoustic model. The effect of this unit is compared with conventional phone units. The dictionary units for language processing are automatically extracted from eojeols appearing in transcriptions. Triphone models are used for acoustic modeling and a trigram model is used for language modeling. Among three major speaker groups in news broadcasts-anchors, journalists and people (those other than anchors or journalists, who are being interviewed), the speech of anchors and journalists, which has a lot of noise, was used for testing and recognition.

• Proceedings of the KSPS conference
• /
• 2003.05a
• /
• pp.75-78
• /
• 2003

In this paper, we describe a cross-morpheme pronunciation variation model which is especially useful for constructing morpheme-based pronunciation lexicon for Korean LVCSR. There are a lot of pronunciation variations occurring at morpheme boundaries in continuous speech. Since phonemic context together with morphological category and morpheme boundary information affect Korean pronunciation variations, we have distinguished pronunciation variation rules according to the locations such as within a morpheme, across a morpheme boundary in a compound noun, across a morpheme boundary in an eojeol, and across an eojeol boundary. In 33K-morpheme Korean CSR experiment, an absolute improvement of 1.16％ in WER from the baseline performance of 23.17％ WER is achieved by modeling cross-morpheme pronunciation variations with a context-dependent multiple pronunciation lexicon.

• MALSORI
• /
• v.55
• /
• pp.103-118
• /
• 2005

In this paper, we describe a morpheme-based pronunciation lexicon useful for Korean LVCSR. The phonemic-context-dependent multiple pronunciation lexicon improves the recognition accuracy when cross-morpheme pronunciation variations are distinguished from within-morpheme pronunciation variations. Since adding all possible pronunciation variants to the lexicon increases the lexicon size and confusability between lexical entries, we have developed a lexicon pruning scheme for optimal selection of pronunciation variants to improve the performance of Korean LVCSR. By building a proposed pronunciation lexicon, an absolute reduction of $0.56\%$ in WER from the baseline performance of $27.39\%$ WER is achieved by cross-morpheme pronunciation variations model with a phonemic-context-dependent multiple pronunciation lexicon. On the best performance, an additional reduction of the lexicon size by $5.36\%$ is achieved from the same lexical entries.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.33-36
• /
• 2002

We implemented a robust Korean broadcast news transcription system for out-of-vocabulary (OOV), tested its performance. The occurrence of OOV words in the input speech is inevitable in large vocabulary continuous speech recognition (LVCSR). The known vocabulary will never be complete due to the existence of for instance neologisms, proper names, and compounds in some languages. The fixed vocabulary and language model of LVCSR system directly face with these OOV words. Therefore our Broadcast news recognition system has an offline OOV update module of language model and vocabulary to solve OOV problem and selects morpheme-based recognition unit (so called, pseudo-morpheme) for OOV robustness.

• MALSORI
• /
• no.62
• /
• pp.133-147
• /
• 2007

In this paper, we introduce a phone vector discrete HMM(PVDHMM) that decodes a phone sequence string, and demonstrates the applicability to spoken document retrieval. The PVDHMM treats a phone recognizer or large vocabulary continuous speech recognizer (LVCSR) as a vector quantizer whose codebook size is equal to the size of its phone set. We apply the PVDHMM to decode the phone sequence strings and compare the outputs with those of a continuous speech recognizer(CSR). Also we carry out spoken document retrieval experiment through PVDHMM word spotter on the phone sequence strings which are generated by phone recognizer or LVCSR and compare its results with those of retrieval through the phone-based vector space model.

• MALSORI
• /
• no.49
• /
• pp.107-121
• /
• 2004

In this paper, we describe a cross-morpheme pronunciation variation model which is especially useful for constructing morpheme-based pronunciation lexicon to improve the performance of a Korean LVCSR. There are a lot of pronunciation variations occurring at morpheme boundaries in continuous speech. Since phonemic context together with morphological category and morpheme boundary information affect Korean pronunciation variations, we have distinguished phonological rules that can be applied to phonemes in within-morpheme and cross-morpheme. The results of 33K-morpheme Korean CSR experiments show that an absolute reduction of 1.45% in WER from the baseline performance of 18.42% WER was achieved by modeling proposed pronunciation variations with a possible multiple context-dependent pronunciation lexicon.

• Proceedings of the KSPS conference
• /
• 2002.11a
• /
• pp.123-126
• /
• 2002

In this paper, we describe our LVCSR system for Korean broadcast news transcription. The main focus is to find the most proper morpheme-based lexical model for Korean broadcast news recognition to deal with the inflectional flexibilities in Korean. There are trade-offs between lexicon size and lexical coverage, and between the length of lexical unit and WER. In our system, we analyzed the training corpus to obtain a small 24k-morpheme-based lexicon with 98.8％ coverage. Then, the lexicon is optimized by combining morphemes using statistics of training corpus under monosyllable constraint or maximum length constraint. In experiments, our system reduced the number of monosyllable morphemes from 52％ to 29％ of the lexicon and obtained 13.24％ WER for anchor and 24.97％ for reporter.

• MALSORI
• /
• no.50
• /
• pp.139-150
• /
• 2004

A Phonetic Tied-Mixture (PTM) model has been proposed as a way of efficient decoding in large vocabulary continuous speech recognition systems (LVCSR). It has been reported that PTM model shows better performance in decoding than triphones by sharing a set of mixture components among states of the same topological location[5]. In this paper we propose a Phonetic Tied-Mixture Syllable (PTMS) model which extends PTM technique up to syllables. The proposed PTMS model shows 13% enhancement in decoding speed than PTM. In spite of difference in context dependent modeling (PTM : cross-word context dependent modeling, PTMS : word-internal left-phone dependent modeling), the proposed model shows just less than 1% degradation in word accuracy than PTM with the same beam width. With a different beam width, it shows better word accuracy than in PTM at the same or higher speed.

• Proceedings of the KSPS conference
• /
• 2006.11a
• /
• pp.24-27
• /
• 2006

This paper describes how a domain dependent pronunciation lexicon is generated and optimized for Korean large vocabulary continuous speech recognition(LVCSR). At the level of lexicon, pronunciation variations are usually modeled by adding pronunciation variants to the lexicon. We propose the criteria for selecting appropriate pronunciation variants in lexicon: (i) likelihood and (ii) frequency factors to select variants. Our experiment is conducted in three steps. First, the variants are generated with knowledge-based rules. Second, we generate a domain dependent lexicon which includes various numbers of pronunciation variants based on the proposed criteria. Finally, the WERs and RTFs are examined with each lexicon. In the experiment, 0.72% WER reduction is obtained by introducing the variants pruning criteria. Furthermore, RTF is not deteriorated although the average number of variants is higher than that of compared lexica.