• The Journal of the Acoustical Society of Korea
• /
• v.21 no.4
• /
• pp.361-373
• /
• 2002

In this paper, we propose a channel normalization method to improve the performance of CMS (cepstral mean subtraction) which is widely adopted to normalize a channel variation for speech and speaker recognition. CMS which estimates the channel effects by averaging long-term cepstrum has a weak point that the estimated channel is biased by the formants of voiced speech which include a useful speech information. The proposed Formant-broadened Cepstral Mean Subtraction (FBCMS) is based on the facts that the formants can be found easily in log spectrum which is transformed from the cepstrum by fourier transform and the formants correspond to the dominant poles of all-pole model which is usually modeled vocal tract. The FBCMS evaluates only poles to be broadened from the log spectrum without polynomial factorization and makes a formant-broadened cepstrum by broadening the bandwidths of formant poles. We can estimate the channel cepstrum effectively by averaging formant-broadened cepstral coefficients. We performed the experiments to compare FBCMS with CMS, PFCMS using 4 simulated telephone channels. In the experiment of channel estimation, we evaluated the distance cepstrum of real channel from the cepstrum of estimated channel and found that we were able to get the mean cepstrum closer to the channel cepstrum due to an softening the bias of mean cepstrum to speech. In the experiment of text-independent speaker identification, we showed the result that the proposed method was superior than the conventional CMS and comparable to the pole-filtered CMS. Consequently, we showed the proposed method was efficiently able to normalize the channel variation based on the conventional CMS.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.4
• /
• pp.224-229
• /
• 2005

In this paper, we Propose a novel cepstrum normalization method which normalizes the probability density function (pdf) of cepstrum for robust speech recognition in additive noise environments. While the conventional methods normalize the first- and/or second-order statistics such as the mean and/or variance of the cepstrum. the proposed method fully normalizes the statistics of cepstrum by making the pdfs of clean and noisy cepstrum identical to each other For the target Pdf, the generalized Gaussian distribution is selected to consider various densities. In recognition phase, we devise a table lookup method to save computational costs. From the speaker-independent isolated-word recognition experiments, we show that the Proposed method gives improved Performance compared with that of the conventional methods, especially in heavy noise environments.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.4
• /
• pp.131-140
• /
• 1996

In this paper, we propose the method that extracts the speech feature using the hearing model through signal precessing techniques. The proposed method includes following procedure ; normalization of the short-time speech block by its maximum value, multi-resolution analysis using the discrete wavelet transformation and re-synthesize using thediscrete inverse wavelet transformation, differentiation after analysis and synthesis, full wave rectification and integration. In order to verify the performance of the proposed speech feature in the speech recognition task, korean digita recognition experiments were carried out using both the dTW and the VQ-HMM. The results showed that, in case of using dTW, the recognition rates were 99.79% and 90.33% for speaker-dependent and speaker-independent task respectively and, in case of using VQ-HMM, the rate were 96.5% and 81.5% respectively. And it indicates that the proposed speech feature has the potentials to use as a simple and efficient feature for recognition task.

• Proceedings of the KSPS conference
• /
• 1996.10a
• /
• pp.435-447
• /
• 1996

In this paper, we propose the method that extracts the speech feature using the hearing model through signal processing techniques. The proposed method includes the following procedure ; normalization of the short-time speech block by its maximum value, multi-resolution analysis using the discrete wavelet transformation and re-synthesize using the discrete inverse wavelet transformation, differentiation after analysis and synthesis, full wave rectification and integration. In order to verify the performance of the proposed speech feature in the speech recognition task, korean digit recognition experiments were carried out using both the DTW and the VQ-HMM. The results showed that, in the case of using DTW, the recognition rates were 99.79% and 90.33% for speaker-dependent and speaker-independent task respectively and, in the case of using VQ-HMM, the rate were 96.5% and 81.5% respectively. And it indicates that the proposed speech feature has the potential for use as a simple and efficient feature for recognition task

• The Journal of the Acoustical Society of Korea
• /
• v.35 no.6
• /
• pp.501-509
• /
• 2016

We propose a method to improve the robustness of speaker verification on short test utterances. The accuracy of the state-of-the-art i-vector/probabilistic linear discriminant analysis systems can be degraded when testing utterance durations are short. The proposed method compensates for utterance variations of short test feature vectors using deep neural networks. We design three different types of DNN (Deep Neural Network) structures which are trained with different target output vectors. Each DNN is trained to minimize the discrepancy between the feed-forwarded output of a given short utterance feature and its original long utterance feature. We use short 2-10 s condition of the NIST (National Institute of Standards Technology, U.S.) 2008 SRE (Speaker Recognition Evaluation) corpus to evaluate the method. The experimental results show that the proposed method reduces the minimum detection cost relative to the baseline system.

• Proceedings of the IEEK Conference
• /
• 2001.09a
• /
• pp.487-490
• /
• 2001

본 논문에서는 기존의 Caussian Mixture Model을 이용한 실시간문맥독립화자인식시스템의 성능을 향상시키기 위하여 화자검증시스템에서 좋은 결과를 나타내는 유사도정규화 ( Likelihood Normalization )방법을 화자식별시스템에 적용하여 시스템을 구현하였으며, 인식실험한 결과에 대해 보고한다. 시스템은 화자모델생성단과 화자식별단으로 구성하였으며, 화자모델생성단에서는, 화자발성의 음향학적 특징을 잘 표현할 수 있는 GMM(Gaussian Mixture Model)을 이용하여 화자모델을 작성하였으며. GMM의 파라미터를 최적화하기 위하여 MLE(Maximum Likelihood Estimation)방법을 사용하였다. 화자식별단에서는 학습된 데이터와 테스트용 데이터로부터 ML(Maximum Likelihood)을 이용하여 프레임단위로 유사도를 계산하였다. 계산된 유사도는 유사도 정규화 과정을 거쳐 스코어( SC)로 표현하였으며, 가장 높은 스코어를 가지는 화자를 인식화자로 결정한다. 화자인식에서 발성의 종류로는 문맥독립 문장을 사용하였다. 인식실험을 위해서는 ETRI445 DB와 KLE452 DB를 사용하였으며. 특징파라미터로서는 켑스트럼계수 및 회귀계수값만을 사용하였다. 인식실험에서는 등록화자의 수를 달리하여 일반적인 화자식별방법과 프레임단위유사도정규화방법으로 각각 인식실험을 하였다. 인식실험결과, 프레임단위유사도정규화방법이 인식화자수가 많아지는 경우에 일반적인 방법보다 향상된 인식률을 얻을수 있었다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.19 no.6
• /
• pp.773-778
• /
• 2009

This paper studied the training methods less affected by the emotional variation for the development of the robust speech recognition system. For this purpose, the effect of emotional variations on the speech signal were studied using speech database containing various emotions. The performance of the speech recognition system trained by using the speech signal containing no emotion is deteriorated if the test speech signal contains the emotions because of the emotional difference between the test and training data. In this study, it is observed that vocal tract length of the speaker is affected by the emotional variation and this effect is one of the reasons that makes the performance of the speech recognition system worse. In this paper, vocal tract normalization method is used to develop the robust speech recognition system for emotional variations. Experimental results from the isolated word recognition using HMM showed that the vocal tract normalization method reduced the error rate of the conventional recognition system by 41.9% when emotional test data was used.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.55-58
• /
• 2000

본 논문에서는 화자 확인 시스템의 등록과 확인 과정의 채널 환경 불일치로 성능이 저하되는 문제를 해결하기 위한 새로운 정규화 방법에 대해 설명한다. 제안된 방법은 첫째, 입력 음성으로부터 효과적으로 채널을 추정$\cdot$보상하고 둘째, 스코어 정규화 과정에서 사칭자 모델로서 사용되는 world모델과의 차이를 채널 추정 및 화자 모델 생성에 효과적으로 사용하는 것을 목표로 한다. 이를 위해 입력 음성의 켑스트럼과 HMM world 모델의 파라메터인 평균 켑스트럼과의 차이를 통해 음소열에 종속적인 채널 켑스트럼인 Phone-Dependent Difference Cepstrum을 추정한다. 한편 입력 음성의 음소열은 world모델의 스코어를 얻는 과정에서 함께 얻어질 수 있다. 채널 추정 실험 결과를 통해서 가장 일반적인 채널 정규화방법인 CMS에 의해 추정된 채널에 비해 실제 채널과 유사하며 화자 고유의 특성을 왜곡시키지 않는 채널 추정이 가능함을 확인할 수 있었다.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.1E
• /
• pp.3-11
• /
• 2002

Broadcast news transcription is one of the hardest tasks in speech recognition because broadcast speech signals have much variability in speech quality, channel and background conditions. We developed a Korean broadcast news speech recognizer. We used a morpheme-based dictionary and a language model to reduce the out-of·vocabulary (OOV) rate. We concatenated the original morpheme pairs of short length or high frequency in order to reduce insertion and deletion errors due to short morphemes. We used a lexicon with multiple pronunciations to reflect inter-morpheme pronunciation variations without severe modification of the search tree. By using the merged morpheme as recognition units, we achieved the OOV rate of 1.7% comparable to European languages with 64k vocabulary. We implemented a hidden Markov model-based recognizer with vocal tract length normalization and online speaker adaptation by maximum likelihood linear regression. Experimental results showed that the recognizer yielded 21.8% morpheme error rate for anchor speech and 31.6% for mostly noisy reporter speech.

• Journal of Institute of Control, Robotics and Systems
• /
• v.16 no.12
• /
• pp.1137-1142
• /
• 2010

This paper studied the speech parameters less affected by the human emotion for the development of the robust speech recognition system. For this purpose, the effect of emotion on the speech recognition system and robust speech parameters of speech recognition system were studied using speech database containing various emotions. In this study, mel-cepstral coefficient, delta-cepstral coefficient, RASTA mel-cepstral coefficient and frequency warped mel-cepstral coefficient were used as feature parameters. And CMS (Cepstral Mean Subtraction) method were used as a signal bias removal technique. Experimental results showed that the HMM based speaker independent word recognizer using vocal tract length normalized mel-cepstral coefficient, its derivatives and CMS as a signal bias removal showed the best performance of 0.78% word error rate. This corresponds to about a 50% word error reduction as compare to the performance of baseline system using mel-cepstral coefficient, its derivatives and CMS.