• Speech Sciences
• /
• v.12 no.3
• /
• pp.125-136
• /
• 2005

Automatic speech recognition is much more difficult in real world. Speech recognition according to SIR (Signal to Interface Ratio) is difficult in situations in which noise of surrounding environment and multi-speaker exists. Therefore, study on main speaker's voice extractions a very important field in speech signal processing in binaural sound. In this paper, we used directional filter and harmonic filter among other existing methods to extract the main speaker's information in binaural sound. The main speaker's voice was extracted using directional filter, and other remaining speaker's information was removed using harmonic filter through main speaker's pitch detection. As a result, voice of the main speaker was enhanced.

• The Journal of the Acoustical Society of Korea
• /
• v.20 no.3
• /
• pp.56-62
• /
• 2001

In this paper, we adopt a missing data theory to speech recognition. It can be used in order to maintain high performance of speech recognizer when the missing data occurs. In general, hidden Markov model (HMM) is used as a stochastic classifier for speech recognition task. Acoustic events are represented by continuous probability density function in continuous density HMM(CDHMM). The missing data theory has an advantage that can be easily applicable to this CDHMM. A marginalization method is used for processing missing data because it has small complexity and is easy to apply to automatic speech recognition (ASR). Also, a spectral subtraction is used for detecting missing data. If the difference between the energy of speech and that of background noise is below given threshold value, we determine that missing has occurred. We propose a new method that examines the reliability of detected missing data using voicing probability. The voicing probability is used to find voiced frames. It is used to process the missing data in voiced region that has more redundant information than consonants. The experimental results showed that our method improves performance than baseline system that uses spectral subtraction method only. In 452 words isolated word recognition experiment, the proposed method using the voicing probability reduced the average word error rate by 12％ in a typical noise situation.

• MALSORI
• /
• no.65
• /
• pp.93-103
• /
• 2008

In this paper, we propose a method for optimizing a multiple pronunciation dictionary used for modeling pronunciation variations of non-native speech. The proposed method removes some confusable pronunciation variants in the dictionary, resulting in a reduced dictionary size and less decoding time for automatic speech recognition (ASR). To this end, a confusability measure is first defined based on the Levenshtein distance between two different pronunciation variants. Then, the number of phonemes for each pronunciation variant is incorporated into the confusability measure to compensate for ASR errors due to words of a shorter length. We investigate the effect of the proposed method on ASR performance, where Korean is selected as the target language and Korean utterances spoken by Chinese native speakers are considered as non-native speech. It is shown from the experiments that an ASR system using the multiple pronunciation dictionary optimized by the proposed method can provide a relative average word error rate reduction of 6.25%, with 11.67% less ASR decoding time, as compared with that using a multiple pronunciation dictionary without the optimization.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.39-43
• /
• 2023

자동 음성 인식(Automatic Speech Recognition, ASR)은 컴퓨터가 인간의 음성을 텍스트로 변환하는 기술이다. 자동 음성 인식 시스템은 다양한 응용 분야에서 사용되며, 음성 명령 및 제어, 음성 검색, 텍스트 트랜스크립션, 자동 음성 번역 등 다양한 작업을 목적으로 한다. 자동 음성 인식의 노력에도 불구하고 노인 음성 인식(Elderly Speech Recognition, ESR)에 대한 어려움은 줄어들지 않고 있다. 본 연구는 노인 음성 인식에 콘포머(Conformer)와 피쳐 퓨전 모듈(Features Fusion Module, FFM)기반 노인 음성 인식 모델을 제안한다. 학습, 평가는 VOTE400(Voide Of The Elderly 400 Hours) 데이터셋으로 한다. 본 연구는 그동안 잘 이뤄지지 않았던 콘포머와 퓨전피쳐를 사용해 노인 음성 인식을 위한 딥러닝 모델을 제시하였다는데 큰 의미가 있다. 또한 콘포머 모델보다 높은 수준의 정확도를 보임으로써 노인 음성 인식을 위한 딥러닝 모델 연구에 기여했다.

• The Journal of the Acoustical Society of Korea
• /
• v.23 no.4E
• /
• pp.112-119
• /
• 2004

This paper presents a method for the automatic recognition of pitch accents with no prior knowledge about the phonetic content of the signal (no knowledge of word or phoneme boundaries or of phoneme labels). The recognition algorithm used in this paper is a time-delay recurrent neural network (TDRNN). A TDRNN is a neural network classier with two different representations of dynamic context: delayed input nodes allow the representation of an explicit trajectory F0(t), while recurrent nodes provide long-term context information that can be used to normalize the input F0 trajectory. Performance of the TDRNN is compared to the performance of a MLP (multi-layer perceptron) and an HMM (Hidden Markov Model) on the same task. The TDRNN shows the correct recognition of $91.9{\%}\;of\;pitch\;events\;and\;91.0{\%}$ of pitch non-events, for an average accuracy of $91.5{\%}$ over both pitch events and non-events. The MLP with contextual input exhibits $85.8{\%},\;85.5{\%},\;and\;85.6{\%}$ recognition accuracy respectively, while the HMM shows the correct recognition of $36.8{\%}\;of\;pitch\;events\;and\;87.3{\%}$ of pitch non-events, for an average accuracy of $62.2{\%}$ over both pitch events and non-events. These results suggest that the TDRNN architecture is useful for the automatic recognition of pitch accents.

• Journal of the Institute of Convergence Signal Processing
• /
• v.7 no.3
• /
• pp.116-121
• /
• 2006

In this paper, we propose the dimension reduction method of multi-dimension speech feature vector for real-time adaptation procedure in various noisy environments. This method which reduces dimensions non-linearly to map the likelihood of speech feature vector and noise feature vector. The LRT(Likelihood Ratio Test) is used for classifying speech and non-speech. The results of implementation are similar to multi-dimensional speech feature vector. The results of speech recognition implementation of detected speech data are also similar to multi-dimensional(10-order dimensional MFCC(Mel-Frequency Cepstral Coefficient)) speech feature vector.

• Phonetics and Speech Sciences
• /
• v.12 no.2
• /
• pp.29-37
• /
• 2020

Recently, the neural network-based deep learning algorithm has dramatically improved performance compared to the classical Gaussian mixture model based hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system. In addition, researches on end-to-end (E2E) speech recognition systems integrating language modeling and decoding processes have been actively conducted to better utilize the advantages of deep learning techniques. In general, E2E ASR systems consist of multiple layers of encoder-decoder structure with attention. Therefore, E2E ASR systems require data with a large amount of speech-text paired data in order to achieve good performance. Obtaining speech-text paired data requires a lot of human labor and time, and is a high barrier to building E2E ASR system. Therefore, there are previous studies that improve the performance of E2E ASR system using relatively small amount of speech-text paired data, but most studies have been conducted by using only speech-only data or text-only data. In this study, we proposed a semi-supervised training method that enables E2E ASR system to perform well in corpus in different domains by using both speech or text only data. The proposed method works effectively by adapting to different domains, showing good performance in the target domain and not degrading much in the source domain.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.12
• /
• pp.17-24
• /
• 2019

The biggest problem in the AI field, which has become a hot topic in recent years, is how to deal with the lack of training data. Since manual data construction takes a lot of time and efforts, it is non-trivial for an individual to easily build the necessary data. On the other hand, automatic data construction needs to handle data quality issue. In this paper, we introduce a method to automatically extract the data required to develop Korean speech command recognizer from the web and to automatically select the data that can be used for training data. In particular, we propose a modified ResNet model that shows modest performance for the automatically constructed Korean speech command data. We conducted an experiment to show the applicability of the command set of the health and daily life domain. In a series of experiments using only automatically constructed data, the accuracy of the health domain was 89.5% in ResNet15 and 82% in ResNet8 in the daily lives domain, respectively.

• Proceedings of the KSPS conference
• /
• 2004.05a
• /
• pp.239-242
• /
• 2004

An automatic pronunciation correction system provides users with correction guidelines for each pronunciation error. For this purpose, we propose a speech recognition system which automatically classifies pronunciation errors when Koreans speak a foreign language. In this paper, we also propose machine scoring methods for automatic assessment of pronunciation quality by the speech recognizer. Scores obtained from an expert human listener are used as the reference to evaluate the different machine scores and to provide targets when training some of algorithms. We use a log-likelihood score and a normalized log-likelihood score as machine scoring methods. Experimental results show that the normalized log-likelihood score had higher correlation with human scores than that obtained using the log-likelihood score.

• MALSORI
• /
• no.45
• /
• pp.35-45
• /
• 2003

Laryngeal diseases cause significant changes in the quality of speech production. Automatic detection of laryngeal diseases by voice is attractive because of its nonintrusive nature. In this paper, we apply speech recognition techniques to detection of laryngeal cancer, and investigate which feature parameters and classification methods are appropriate for this purpose. Linear Predictive Cepstral Coefficients (LPCC) and Mel-Frequency Cepstral Coefficients (MFCC) are examined as feature parameters, and parameters reflecting the periodicity of speech and its perturbation are also considered. As for classifier, multilayer perceptron neural networks and Gaussian Mixture Models (GMM) are employed. According to our experiments, higher order LPCC with the periodic information parameters yields the best performance.