• Proceedings of the KSPS conference
• /
• 2005.04a
• /
• pp.91-94
• /
• 2005

The performance of speech recognition is degraded by the mismatch between training and test environments. Many methods have been presented to compensate for additive noise and channel effect in the cepstral domain, and Cepstral Mean Subtraction (CMS) is the representative method among them. Recently, high order cepstral moment normalization method has introduced to improve recognition accuracy. In this paper, we apply high order moment normalization method and smoothing filter for real-time processing. In experiments using Aurora2 DB, we obtained error rate reduction of 49.7% with the proposed algorithm in comparison with baseline system.

• MALSORI
• /
• no.60
• /
• pp.191-201
• /
• 2006

Discrimination between speech and music is important in many multimedia applications. Previously we proposed a new parameter for speech/music discrimination, the mean of minimum cepstral distances (MMCD), and it outperformed the conventional parameters. One weakness of MMCD is that its performance depends on range of candidate frames to compute the minimum cepstral distance, which requires the optimal selection of the range experimentally. In this paper, to alleviate the problem, we propose a multi-dimensional MMCD parameter which consists of multiple MMCDS with combination of different candidate frame ranges. Experimental results show that the multi-dimensional MMCD parameter yields an error rate reduction of 22.5% compared with the optimally chosen one-dimensional MMCD parameter.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1996.10a
• /
• pp.56-60
• /
• 1996

본 논문에서 핵심어 검출(Keyword spotting ) 시스템에서의 채널 왜곡에 대한 보상방법등의 성능을 비교하였다. 훈련을 음성과 인식실험용 음성은 서로 다른 환경에서 수집되었으며, 특별히 인식실험용 음성으로는 전화망을 통한 음성 데이터를 이용하였다. 전화망을 통한 음성인식에서는 채널왜곡과 부가잡음에 의해서 음성신호에 왜곡이 생기므로 이들에 대한 적적한 보상이 필요하다. 본 논문에서는 채널 왜곡보상을 위한 처리방법으로 널리 사용되고 있는 global cepstral mean substraction (GCMS), local cepstral mean subtraction(LCMS) 그리고 RASTA processing을 적용하였다. 그리고 인식성능의 개선을 위해 이들 방법을 likelihood ration scorning 에 의한 후처리 과정을 적용하였다. 인식실험결과 이들 방법 모두 채널왜곡 보상을 하지 않았을 경우와 비교하여 더 좋은 인식성능을 얻을 수 있었으며, 그 중 후처리를 적용한 LCMS 방법이 가장 우수한 성능을 나타내었다.

• The Journal of the Acoustical Society of Korea
• /
• v.7 no.4
• /
• pp.77-85
• /
• 1988

Many kinds of conditioning monitoring technique have been studied, so this study has inverstigated the possibility of checking the trend in the fault diagnosis of ball bearing, one of the elements of rotating machine, by applying the cepstral analyisis method using the adaptive noise cancelling (ANC) method. And computer simulation is conducted in order to verify the usefulness of ANC. The optimal adaptation gain in adaptive filter is estimated, the performance of ANC according to the change of the signal to noise ratio and convergence of least mean square algorithm is considered by simulation. It is verified that cepstral analysis using ANC method is more effective than the conventional cepstral analysis method in bearing fault diagnosis.

• The Journal of the Acoustical Society of Korea
• /
• v.22 no.3
• /
• pp.182-187
• /
• 2003

In this thesis, we propose the algorithm to improve the performance of speaker verification that is mapping feature parameters by using RBF neural network. There is a big difference between wire vector region and wireless one which comes from the same speaker. For wire/wireless speakers model production, speaker verification system should distinguish the wire/wireless channel that based on speech recognition system. And the feature vector of untrained channel models is mapped to the feature vector(LPC Cepstrum) of trained channel model by using RBF neural network. As a simulation result, the proposed algorithm makes 0.6%∼10.5% performance improvement compared to conventional method such as cepstral mean subtraction.

• Journal of the Institute of Electronics Engineers of Korea TE
• /
• v.39 no.2
• /
• pp.60-65
• /
• 2002

In this paper, we propose GFCC(gammatone filter frequency cepstrum coefficient) parameter which was based on the auditory characteristic for accomplishing better speech recognition rate. And it is performed the experiment of speech recognition for isolated word acquired from telephone network. For the purpose of comparing GFCC parameter with other parameter, the experiment of speech recognition are carried out using MFCC and LPCC parameter. Also, for each parameter, we are implemented CMS(cepstral mean subtraction)which was applied or not in order to compensate channel distortion in telephone network. Accordingly, we found that the recognition rate using GFCC parameter is better than other parameter in the experimental result.

• The Journal of the Acoustical Society of Korea
• /
• v.33 no.1
• /
• pp.68-74
• /
• 2014

In a previous work, we proposed a novel approach of histogram equalization using a supplement set which is composed of centroids of Fuzzy C-Means of the background utterances. The performance of the proposed method is affected by the size of the supplement set, but it is difficult to find the best size at the point of recognition. In this paper, we propose a histogram equalization using a supplement set for majority voting based speaker identification. The proposed method identifies test utterances using a majority voting on the histogram equalization methods with various sizes of supplement sets. The proposed method is compared with the conventional feature normalization methods such as CMN(Cepstral Mean Normalization), MVN(Mean and Variance Normalization), and HEQ(Histogram Equalization) and the histogram equalization method using a supplement set.

• MALSORI
• /
• no.64
• /
• pp.155-169
• /
• 2007

Discrimination between speech and music is important in many multimedia applications. In our previous work, focusing on the spectral change characteristics of speech and music, we presented a method using the mean of minimum cepstral distances (MMCD), and it showed a very high discrimination performance. In this paper, to further improve the performance, we propose to employ time-averaged MFCC in computing the MMCD. Our experimental results show that the proposed method enhances the discrimination between speech and music. Moreover, the proposed method overcomes the weakness of the conventional MMCD method whose performance is relatively sensitive to the choice of the frame interval to compute the MMCD.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.1E
• /
• pp.32-36
• /
• 1999

In the area of the speech synthesis techniques, the waveform coding methods maintain the intelligibility and naturalness of synthetic speech. In order to apply the waveform coding techniques to synthesis by rule, we must be able to alter the pitches of synthetic speech. In this paper, we propose a new pitch altering method that compensates phase distortion of the cepstral pitch alteration method with time scaling method in the time domain. This method can remove some spectrum distortion which is occurred in conjunction point between the waveforms. For performance test the spectrum distortion rate was used as objective criterion and the MOS(Mean Opinion Score) was used as subjective criterion. As a result, the spectrum distortion and MOS are obtained by 0.66% and 3.9, respectively.

• Speech Sciences
• /
• v.10 no.3
• /
• pp.263-277
• /
• 2003

One of major problems in speech recognition is performance degradation due to the mismatch between the training and test environments. Recently, Stereo-based Piecewise LInear Compensation for Environments (SPLICE), which is frame-based bias removal algorithm for cepstral enhancement using stereo training data and noisy speech model as a mixture of Gaussians, was proposed and showed good performance in noisy environments. In this paper, we propose several methods to improve the conventional SPLICE. First we apply Cepstral Mean Subtraction (CMS) as a preprocessor to SPLICE, instead of applying it as a postprocessor. Secondly, to compensate residual distortion after SPLICE processing, two-stage SPLICE is proposed. Thirdly we employ phonetic information for training SPLICE model. According to experiments on the Aurora 2 database, proposed method outperformed the conventional SPLICE and we achieved a 50% decrease in word error rate over the Aurora baseline system.