• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.29B no.2
• /
• pp.50-59
• /
• 1992

In the most studies of Korean character recognition so far, they first classify the characters to 6 types according to their structures and then recognize the characters by identifying their basic components named $'$jaso.$'$ In the study, we propose a method which recognizes the characters without using structure types and is applied to reading documents containing both Korean and Chinese characters. We first classify Korean and Chinese characters by using a modified SOFM model. Then we recognize the characters in each class by using an APC neural network which has the advantage of fast leaning speed and the capablity of additive learning. An experimental result demonstrated the usefulness of the approach with the recognition rate of $\%.$\%.

• 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.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.30B no.9
• /
• pp.84-93
• /
• 1993

Recently, several nonlinear shape normalization methods have been proposed in order to compensate for the shape distortions in handwritten characters. In this paper, we review these nonlinear shape normalization methods from the two points of view : feature projection and feature density equalization. The former makes feature projection histogram by projecting a certain feature at each point of input image into horizontal-or vertical-axis and the latter equalizes the feature densities of input image by re-sampling the feature projection histogram. A systematic comparison of these methods has been made based on the following criteria: recognition rate, processing speed, computational complexity and measure of variation. Then, we present the result of quantitative evaluation of each method based on these criteria for a large variety of handwritten Hangul syllables.

• Phonetics and Speech Sciences
• /
• v.6 no.1
• /
• pp.77-83
• /
• 2014

This paper proposes a speaker and environment clustering method in order to overcome the degradation of the speech recognition performance caused by various noise and speaker characteristics. In this paper, instead of using the distance between Gaussian mixture model (GMM) weight vectors as in the Google's approach, the distance between the adapted mean vectors based on the modified maximum a posteriori (MAP) adaptation is used as a distance measure for vector quantization (VQ) clustering. According to our experiments on the simulation data generated by adding noise to clean speech, the proposed clustering method yields error rate reduction of 10.6% compared with baseline speaker-independent (SI) model, which is slightly better performance than the Google's approach.

• MALSORI
• /
• no.46
• /
• pp.117-126
• /
• 2003

This paper presents our language model adaptation for Korean spontaneous speech recognition. Korean spontaneous speech is observed various characteristics of content and style such as filled pauses, word omission, and contraction as compared with the written text corpus. Our approaches focus on improving the estimation of domain-dependent n-gram models by relevance weighting out-of-domain text data, where style is represented by n-gram based tf/sup */idf similarity. In addition to relevance weighting, we use disfluencies as Predictor to the neighboring words. The best result reduces 9.7% word error rate relatively and shows that n-gram based relevance weighting reflects style difference greatly and disfluencies are good predictor also.

• Journal of Power System Engineering
• /
• v.4 no.2
• /
• pp.25-30
• /
• 2000

The grinding process is very complex and relates many parameters to control the process. As this reason, a theoretical analysis and a quantitative estimation of the grinding process has not been well established. In this study, the in-process monitoring system was suggested by applying the neural network for monitoring and shooting the malfunction of cylindrical plunge grinding process. This system used the power signals from the electric power meter. This neural network was composed of processing elements [4-(5-5)-3] with 4 identified power parameters. Because sensitivity is blunted some minute vibration components, the simulation result of this system has appeared about 10% erroneous recognition in the uncertain pattern and the average success rate of the trouble recognition was about 90%. Consequently, the developed system, which applied to the power signals, can be recognize enough to monitor the grinding process as in-process.

• ETRI Journal
• /
• v.32 no.5
• /
• pp.810-818
• /
• 2010

This paper presents a new framework for integrating untranscribed spoken content into the acoustic training of an automatic speech recognition system. Untranscribed spoken content plays a very important role for under-resourced languages because the production of manually transcribed speech databases still represents a very expensive and time-consuming task. We proposed two new methods as part of the training framework. The first method focuses on combining initial acoustic models using a data-driven metric. The second method proposes an improved acoustic training procedure based on unsupervised transcriptions, in which word endings were modified by broad phonetic classes. The training framework was applied to baseline acoustic models using untranscribed spoken content from parliamentary debates. We include three types of acoustic models in the evaluation: baseline, reference content, and framework content models. The best overall result of 18.02% word error rate was achieved with the third type. This result demonstrates statistically significant improvement over the baseline and reference acoustic models.

• MALSORI
• /
• no.56
• /
• pp.207-223
• /
• 2005

The goal of language model adaptation is to improve the background language model with a relatively small adaptation corpus. This study presents a language model adaptation technique where additional text data for the adaptation do not exist. We propose the information retrieval (IR) technique with N-gram language modeling to collect the adaptation corpus from baseline text data. We also propose to use a dynamic language model interpolation coefficient to combine the background language model and the adapted language model. The interpolation coefficient is estimated from the word hypotheses obtained by segmenting the input speech data reserved for held-out validation data. This allows the final adapted model to improve the performance of the background model consistently The proposed approach reduces the word error rate by $13.6\%$ relative to baseline 4-gram for two-hour broadcast news speech recognition.

• Journal of Korea Multimedia Society
• /
• v.8 no.1
• /
• pp.56-61
• /
• 2005

In this thesis, propose speech period and pitch detection for continuous speech recognition. This mathod is distinguishes between vowel and consonant to frame unit in continuous speech, for distinguishable voice. Powerful extraction of speech period could threshold energy make use of input signal to real noise environment. Also algorithm of this method distinguish between vowel and consonant at the same time in voice make use of zero crossing rate and short time energy to extractible speech period.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.5
• /
• pp.58-62
• /
• 1999

This paper proposes use of acoustic parameters to improve the discriminability among digit models in Korean connected digit recognition. The proposed method used the logarithmic values of energy ratio between the predetermined frequency bands as additional feature parameters, based on the acoustic-phonetic knowledge. The results of our experiment show that the proposed method reduced the error rate by 46% in comparison with the baseline system. And incorporation of channel compensation technique in the proposed method yielded error reduction of about 69%.