• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.6
• /
• pp.279-283
• /
• 2004

In general, triangular shape filters are used in the filter bank when we extract MFCC feature parameters from the spectrum of the speech signal. A different approach, which uses specific filter shapes in the filter bank that are optimized to the spectrum of training speech data, is proposed by Lee et al. to improve the recognition rate. A principal component analysis method is used to get the optimized filter coefficients. Using a large amount of 4-digit telephone speech database, in this paper, we get the MFCCs based on the PCA-optimized filter bank and compare the recognition performance with conventional MFCCs and direct weighted filter bank based MFCCs. Experimental results have shown that the MFCC based on the PCA-optimized filter bank give slight improvement in recognition rate compared to the conventional MFCCs but fail to achieve better performance than the MFCCs based on the direct weighted filter bank analysis. Experimental results are discussed with our findings.

• Journal of the Korea Society of Computer and Information
• /
• v.28 no.1
• /
• pp.87-92
• /
• 2023

With the continuous development of the speech recognition system, the recognition rate for speech has developed rapidly, but it has a disadvantage in that it cannot accurately recognize the voice due to the noise generated by mixing various voices with the noise in the use environment. In order to increase the vocabulary recognition rate when processing speech with environmental noise, noise must be removed. Even in the existing HMM, CHMM, GMM, and DNN applied with AI models, unexpected noise occurs or quantization noise is basically added to the digital signal. When this happens, the source signal is altered or corrupted, which lowers the recognition rate. To solve this problem, each voice In order to efficiently extract the features of the speech signal for the frame, the MFCC was improved and processed. To remove the noise from the speech signal, the noise removal method using the Gaussian model applied noise deviation estimation was improved and applied. The performance evaluation of the proposed model was processed using a cross-correlation coefficient to evaluate the accuracy of speech. As a result of evaluating the recognition rate of the proposed method, it was confirmed that the difference in the average value of the correlation coefficient was improved by 0.53 dB.

• Annual Conference on Human and Language Technology
• /
• 2006.10e
• /
• pp.171-176
• /
• 2006

본 논문에서는 n-best 리랭킹을 이용한 한-영 통계적 음성 번역 시스템에 대해 논하고 있다. 보통의 음성 번역 시스템은 음성 인식 시스템, 자동 번역 시스템, 음성 합성 시스템이 순차적으로 결합되어 있다. 하지만 본 시스템은 음성 인식 오류에 보다 강인한 시스템을 만들기 위해 음성 인식 시스템으로부터 n-best 인식 문장을 추출하여 번역 결과와 함께 리랭킹의 과정을 거친다. 자동 번역 시스템으로 구절기반 통계적 자동 번역 모델을 사용하여, 음성 인식기의 발음 모델에서 기본 단어 단위와 맞추어 번역 모델과 언어 모델을 훈련시킴으로써 음성 번역 시스템에서 형태소 분석기를 제거할 수 있다. 또한 음성 인식 시스템에서 상황 별로 언어 모델을 분리하여 처리함으로써 자동 번역 시스템에 비해 부족한 음성 인식 시스템의 처리 범위를 보완할 수 있었다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2004.05a
• /
• pp.793-796
• /
• 2004

본 논문은 의료분야의 음성 인식 시스템의 발음모델의 성능 향상에 필요한 올바른 영어 단어 발음열을 얻고자 한다. 본 시스템의 텍스트는 의료 전문 용어인 영어와 한글의 조합으로 되어있어, 한국어 G2P 성능뿐만 아니라 영어 G2P의 성능도 중요한 문제가 되고 있다. 또한 본 시스템의 의료 음성 데이터가 한국 화자로써, 표기열이 영어식 발음 폰셋으로 변환되면 효율적이지 못하다. 이를 위해, 영어 G2P의 결과를 한국 화자에 적합하게 변환해주는 방법론이 필요하게 된다. 따라서 본 논문에서 제안하는 방법은 음성 텍스트에서 영어만 추출한 후, 영어 G2P 프로그램(addttp, NIST)을 이용해 발음열을 구한다. 그리고 한국 화자의 실제 음성을 통해 얻은 정답 발음열을 구하여 서로 비교한다. 비교를 위해 각 발음열의 한 폰씩 정렬을 수행한 후, 삽입, 삭제, 대치 에러가 이러나는 쌍과 좌우 바이그램 정보를 추출한다. 마지막으로, 좌우 바이그램 정보에서 best1의 에러 패턴을 통해 모든 단어에 적용한다. 이 때, 최종적으로 실보다 득이 되는 에러패턴만을 추출, 적용한다. 실험에서는 26여개의 에러 패턴을 찾을 수 있어, 8%의 올바른 발음열을 추가적으로 얻는데 성공하였다.

• The Journal of the Acoustical Society of Korea
• /
• v.29 no.1
• /
• pp.82-90
• /
• 2010

This paper proposes an efficient method to enhance the quality of synthesized speech in HMM based speech synthesis system. The proposed method trains spectral parameters and excitation signals using Gaussian mixture model, and estimates appropriate excitation signals from spectral parameters during the synthesis stage. Both WB-PESQ and MUSHRA results show that the proposed method provides better speech quality than conventional HMM based speech synthesis system.

• Journal of Broadcast Engineering
• /
• v.17 no.3
• /
• pp.519-528
• /
• 2012

Voice activity detection (VAD) is generally conducted by extracting features from the acoustic signal and a decision rule. The performance of such VAD algorithms driven by the input acoustic signal highly depends on the acoustic noise. When video signals are available as well, the performance of VAD can be enhanced by using the visual information which is not affected by the acoustic noise. Previous visual VAD algorithms usually use single visual feature to detect the lip activity, such as active appearance models, optical flow or intensity variation. Based on the analysis of the weakness of each feature, we propose to combine intensity change measure and the optical flow in the mouth region, which can compensate for each other's weakness. In order to minimize the computational complexity, we develop simple measures that avoid statistical estimation or modeling. Specifically, the optical flow is the averaged motion vector of some grid regions and the intensity variation is detected by simple thresholding. To extract the mouth region, we propose a simple algorithm which first detects two eyes and uses the profile of intensity to detect the center of mouth. Experiments show that the proposed combination of two simple measures show higher detection rates for the given false positive rate than the methods that use a single feature.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.326-334
• /
• 2022

In this paper, we propose speech enhancement algorithm as a pre-processing for robust speech recognition in noisy environments. Auxiliary-function-based Independent Vector Analysis (AuxIVA) is performed with weighted covariance matrix using time-varying variances with scaling factor from target masks representing time-frequency contributions of target speech. The mask estimates can be obtained using Neural Network (NN) pre-trained for speech extraction or diffuseness using Coherence-to-Diffuse power Ratio (CDR) to find the direct sounds component of a target speech. In addition, outputs for omni-directional noise are closely chained by sharing the time-varying variances similarly to independent subspace analysis or IVA. The speech extraction method based on AuxIVA is also performed in Independent Low-Rank Matrix Analysis (ILRMA) framework by extending the Non-negative Matrix Factorization (NMF) for noise outputs to Non-negative Tensor Factorization (NTF) to maintain the inter-channel dependency in noise output channels. Experimental results on the CHiME-4 datasets demonstrate the effectiveness of the presented algorithms.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.7
• /
• pp.1307-1311
• /
• 2006

In this paper, we propose new speech feature parameter using the Matrix Factorization for appearance part-based features of speech spectrum. The proposed parameter represents effective dimensional reduced data from multi-dimensional feature data through matrix factorization procedure under all of the matrix elements are the non-negative constraint. Reduced feature data presents p art-based features of input data. We verify about usefulness of NMF(Non-Negative Matrix Factorization) algorithm for speech feature extraction applying feature parameter that is got using NMF in Mel-scaled filter bank output. According to recognition experiment results, we confirm that proposed feature parameter is superior to MFCC(Mel-Frequency Cepstral Coefficient) in recognition performance that is used generally.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.10 no.1
• /
• pp.51-55
• /
• 2010

Tele-marketing service has been provided through voice media in a several places such as modern call centers. In modern call centers, they are trying to measure their service quality, and one of the measuring method is a extracting speaker's feeling information in their voice. In this study, it is proposed to analyze speaker's voice in order to extract their feeling information. For this purpose, a person's feeling is categorized by analyzing several types of signal parameters in the voice signal. A person's feeling can be categorized in four different states: joy, sorrow, excitement, and normality. In a normal condition, excited or angry state can be major factor of service quality. In this paper, it is proposed to select a conversation with problems by extracting the speaker's feeling information based on pitches and amplitudes of voice.

• Journal of KIISE:Software and Applications
• /
• v.31 no.6
• /
• pp.799-807
• /
• 2004

In this paper, we deal with an emotion recognition method using facial images and speech signal. Six basic human emotions including happiness, sadness, anger, surprise, fear and dislike are investigated. Emotion recognition using the facial expression is performed by using a multi-resolution analysis based on the discrete wavelet transform. And then, the feature vectors are extracted from the linear discriminant analysis method. On the other hand, the emotion recognition from speech signal method has a structure of performing the recognition algorithm independently for each wavelet subband and then the final recognition is obtained from a multi-decision making scheme.