• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1997.10a
• /
• pp.291-295
• /
• 1997

현재에 주로 개발되어 상용화가 시작되고 있는 음성인식 시스템의 대부분은 단어인식을 기분으로 하는 시스템으로 적용 단어수를 늘려줌으로서 인식범위를 늘일 수 있으나, 그에 따라 검색해야하는 단어수가 늘어남으로서 전체적인 시스템의 속도 및 성능이 저하되는 경향이 있다. 이러한 단점의 극복을 위하여 본 논문에서는 HMM(Hidden Markov Model)과 GA(Genetic Algorithm)를 이용한 한국어 음성의 음소단위 인식 시스템을 구현하였다. 음성 특징으로는 LPC Cepstrum 계수를 사용하였으며, 인식시는 인식대상이 되는 단어에 대하여 GA(Genetic Algorithm)을 통하여 각 음소를 분리하고, 음소단위로 학습된 HMM 파라미터를 적용하여 인식함으로써 각각의 음소별 가능하도록 하는 방법을 제안하였다.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.20 no.11
• /
• pp.1009-1017
• /
• 2010

As rotating machines play an important role in industrial applications such as aeronautical, naval and automotive industries, many researchers have developed various condition monitoring system and fault diagnosis system by applying various techniques such as signal processing and pattern recognition. Recently, fault diagnosis systems using artificial neural network have been proposed. For effective fault diagnosis, this paper used MLP(multi-layer perceptron) network which is widely used in pattern classification. Since using obtained signals without preprocessing as inputs of neural network can decrease performance of fault classification, it is very important to extract significant features of captured signals and to apply suitable features into diagnosis system according to the kinds of obtained signals. Therefore, this paper proposes the decision method of the proper feature vectors about each fault signal for neural-network-based fault diagnosis system. We applied LPC coefficients, maximum magnitudes of each spectral section in FFT and RMS(root mean square) and variance of wavelet coefficients as feature vectors and selected appropriate feature vectors as comparing error ratios of fault diagnosis for sound, vibration and current fault signals. From experiment results, LPC coefficients and maximum magnitudes of each spectral section showed 100 % diagnosis ratios for each fault and the method using wavelet coefficients had noise-robust characteristic.

• The Journal of the Acoustical Society of Korea
• /
• v.10 no.4
• /
• pp.5-11
• /
• 1991

This paper suggests a segmentation method of speech signal into phonemic units. The suggested segmentation system is speaker-independent and performed without anyprior information of speech signal. In segmentation process, we first divide input speech signal into purevoiced region and not pure voiced speech regions. After then we apply the second algorithm which segments each region into the detailed phonemic units by using the voiced detection parameters, i.e., the time variation of 0th LPC cepstrum coefficient parameter and the ZCR parameter. Types of speech, used to prove the availability of segmentation algorithm suggested in this paper, are the vocabulary composed of isolated words and continuous words. According to the experiments, the successful segmentation rate for 507 phonemic units involved in the total vocabulary is 91.7%.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.13 no.5
• /
• pp.576-581
• /
• 2003

Speech signal has the features of speakers. In this paper, we propose the speaker identification system which use the incremental learning based on neural network. Recorded speech signal through the Mic is passed the end detection and is divided voiced signal and unvoiced signal. The extracted 12 order cpestrum are used the input data for neural network. Incremental learning is the learning algorithm that the learned weights are remembered and only the new weights, that is created as adding new speaker, are trained. The architecture of neural network is extended with the number of speakers. So, this system can learn without the restricted number of speakers.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.32 no.3C
• /
• pp.234-240
• /
• 2007

In this paper, low complexity block constrained trellis coded quantization (BC-TCQ) structures are introduced, and a predictive BC TCQ encoding method is developed for quantization of line spectrum frequencies (LSF) parameters for narrowband speech coding applications. Trellis-coded quantization(TCQ) is a form of VQ that builds the VQ codebook from interleaved constituent scalar quantization codebooks. The performance is compared to the other VQ, demonstrating reduction in spectral distortion and significant reduction in encoding complexity. The predictive BC-TCQ is about 0.47107 dB superior to the IS-641 split-VQ, 26bits/frame, in spectral distortion sense. The BC-TCQ is 64.54%, 76.93%, 2.35% of the IS-641 split-VQ, respectively, in the complexity of the additions, multiplies, comparisons.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1985.10a
• /
• pp.75-78
• /
• 1985

본 연구는 시간 영역의 parament를 이용하여 한국어 숫자음(영, 일, 이, 삼, 사, 오, 육, 칠, 팔, 구)을 인식했다. 입력 음성 신호 X(n)의 Beginning Point와 Ending point를 ZCR(Zero-crossing Rate), Magnitude, Energy, Autocorrelation을 이용 Beginning point와 Ending point를 구하고 자음부의 인식은 위 계수들을 이용하여 행했다. 또, 유성음 부분에서는 PARCOR(Partial Autocorrelation), LPC(Linear Predictive Coding)를 이용 모음부와 유성자음을 인식하여 모음을 6개 부류(ㅏ, ㅑ, ㅗ, ㅜ, ㅠ, ㅣ)로 구분 인식했다. 이 방법에 의하면 입력 음성 신호 X(n)의 B.P(Beginning Point)와 E.P(Ending Point)를 쉽게 추출 가능하며 또한 각 Parameter를 이용하여 94.4%의 인식율을 얻었다.

• The Journal of the Acoustical Society of Korea
• /
• v.9 no.6
• /
• pp.15-21
• /
• 1990

유-무-묵음 분류 과정은 음성분석시에 아주 중요한 문제중의 하나이다. 음성에너지, ZCR, 자기 상관계수, LPC 계수, 예측에러 에너지등을 퍼래미터로 사용하여 지금까지 많은 분류기법이 제안되어져 왔다. 이런기법들은 기본적으로 퍼래미터를 추출해야 하고, 이 때문에 많은 계산량이 요구되고, 이들 퍼 래미터는 음성 본래의 정보들의 대부분을 상실하게 된다. 이 때문에 각 프레임의 진폭분포를 사용하는 새로운 앨고리즘을 제안하였다. 첫째로 V-U-S 영역은 개별 진폭분포형태를 가지기 때문에 주어진 프레 임에서 진폭분포를 구한다. 그런 다음에는 신경망을 통해 분류를 하게 된다. 신경망은 문덕값을 별도로 선정할 필요없고, 배경잡음에 강력하며, 또한 실시간 처리에 적합하다.

• The Journal of the Acoustical Society of Korea
• /
• v.6 no.1
• /
• pp.23-28
• /
• 1987

This paper discusses a reference pattern creation for a speaker-independent Korean isolated word by using the clustering. Tn this paper we permitted to top 3 clusters and created reference pattern by Minimax Criterion. The features parameter used the LPC Coefficients and Autocorrelation and simple Itakura distance measure was used to measure similarity between patterns. With word reference patterns obtained as described above the recognition rate was within one choice only $55.9\%$, two choice only $76.9\%$, three choice only $89.5\%$.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.3 no.4
• /
• pp.23-27
• /
• 2010

This paper performs emotion classification test to find out the best parameter of electroencyphalogram(EEG) signal. Linear predictor coefficients, band cross-correlation coefficients of fast Fourier transform(FFT) and autoregressive model spectra are used as the parameters of 10-channel EEG signal. A multi-layer neural network is used as the pattern classifier. Four emotions for relaxation, joy, sadness, irritation are induced by four university students of an acting circle. Electrode positions are Fp1, Fp2, F3, F4, T3, T4, P3, P4, O1, O2. As a result, the Linear predictor coefficients showed the best performance.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.40 no.1
• /
• pp.74-81
• /
• 2003

In this study, we'd implemented a real time embedded speech recognition system that requires minimum memory size for speech recognition engine and DB. The word to be recognized consist of 40 commands used in a PCS phone and 10 digits. The speech data spoken by 15 male and 15 female speakers was recorded and analyzed by short time analysis method, which window size is 256. The LPC parameters of each frame were computed through Levinson-Burbin algorithm and they were transformed to Cepstrum parameters. Before the analysis, speech data should be processed by pre-emphasis that will remove the DC component in speech and emphasize high frequency band. Baum-Welch reestimation algorithm was used for the training of HMM. In test phone, we could get a recognition rate using likelihood method. We implemented an embedded system by porting the speech recognition engine on ARM core evaluation board. The overall recognition rate of this system was 95%, while the rate on 40 commands was 96% and that 10 digits was 94%.