• IEIE Transactions on Smart Processing and Computing
• /
• 제3권5호
• /
• pp.259-266
• /
• 2014

In this paper, we propose a novel technique for noise robust automatic speech recognition (ASR). The development of ASR techniques has made it possible to recognize isolated words with a near perfect word recognition rate. However, in a highly noisy environment, a distinct mismatch between the trained speech and the test data results in a significantly degraded word recognition rate (WRA). Unlike conventional ASR systems employing Mel-frequency cepstral coefficients (MFCCs) and a hidden Markov model (HMM), this study employ histogram of oriented gradient (HOG) features and a Support Vector Machine (SVM) to ASR tasks to overcome this problem. Our proposed ASR system is less vulnerable to external interference noise, and achieves a higher WRA compared to a conventional ASR system equipped with MFCCs and an HMM. The performance of our proposed ASR system was evaluated using a phonetically balanced word (PBW) set mixed with artificially added noise.

• 한국음향학회:학술대회논문집
• /
• 한국음향학회 1994년도 제11회 음성통신 및 신호처리 워크샵 논문집 (SCAS 11권 1호)
• /
• pp.395-398
• /
• 1994

In this paper the word recognition using sub-model based Hidden Markov Model was studied. Phoneme models were composed of 61 phonemes in therms of Korean language pronunciation characteristic. Using this, word model was maded by serial concatenation. But, in case of this phoneme concatenation, the second and the third phoneme of syllable are overlapped in distribution at the same time. So considering this, the method that combines the second and the third phoneme to one model was proposed. And to prevent the increase in number of model, similar phonemes were combined to one, and finially, 57 models were created. In experiment proper model structure of sub-model was searched for, and recognition results were compared. So similar recognition results were maded, and overall recognition rates were increased in case of using parameter tying method.

• 제어로봇시스템학회논문지
• /
• 제16권12호
• /
• 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.

• 한국통신학회논문지
• /
• 제17권11호
• /
• pp.1199-1205
• /
• 1992

본 논문에서는 HMM(Hidden Markov Models)에서 문제점이 되는 개인차에의한 변동을 흡수하고, 적은 학습 데이타로서 인식률을 향상시키기 위하여 신경회로망을 이용한 NN-HMM(Neural Network Hidden Makov Models)에 의해 한국어 인식에 관하여 연구하였다. 이 방법은 HMM과 신경회로망의 출력을 각각 독립적인 인식값으로 가정하여 두 시스템의 확률곱으로 서로 보정되어 최대 인식확률의 음성모델을 인식하는 음성인식 시스템이다. 본 방법의 타당성을 평가하기 위하여 남, 여화자가 28개의 DDD 지역명을 발성한 음성데이타로 실험한 결과, 이산분포 HMM에 의한 방법에서는 91[%], 신경회로망에 의한 방법에서는 89[%], 제안된 방법에서는 95[%]의 향상된 인식률을 얻으므로써 인식성능의 우수함을 확인하였다.

• 한국통신학회논문지
• /
• 제17권1호
• /
• pp.20-28
• /
• 1992

조음결합은 연결단어 음성인식에서 오인식을 발생시키는 주된 요인이다. 본 연구에서는 표준패턴의 접속 부분에서 선행단어의 영향을 받아 후속단어의 선두 부분하는 것을 고려해 조음결합 모델을 적용하여 후속단어에 변형을 주었다. 조음결합 모델은 임계제동형 선형 2차계를 이용하였으며 1단 DP매칭 인식 알고리즘에 이 모델을 조합하여 효과를 조사했다. 그 결과 화자 5명에 의한 4연 숫자음 35어의 인식 실험에서 조음결합 처리에 의한 인식율이 평균 4.7%개선 되었다.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1988년도 전기.전자공학 학술대회 논문집
• /
• pp.7-9
• /
• 1988

K-means iteration method is generally used for creating the templates in speaker-independent isolated-word recognition system. In this paper the initialization method of initial centers is proposed. The concepts are sorting and trace segmentation. All the tokens are sorted and segmented by trace segmentation so that initial centers are decided. The performance of this method is evaluated by isolated-word recognition of Korean digits. The highest recognition rate is 97.6%.

• 한국인지과학회:학술대회논문집
• /
• 한국인지과학회 2002년도 춘계학술대회
• /
• pp.111-116
• /
• 2002

The regularity effect for printed word recognition and naming depends on ambiguities between single letters (small grain-size) and their phonemic values. As a given word is repeated and becomes more familiar, letter-aggregate size (grain-size) is predicted to increase, thereby decreasing the ambiguity between spelling pattern and phonological representation and, therefore, decreasing the regularity effect. Lexical decision and naming tasks studied the effect of repetition on the regularity effect for words. The familiarity of a word from was manipulated by presenting low and high frequency words as well as by presenting half the stimuli in mixed upper- and lowercase letters (an unfamiliar form) and half in uniform case. In lexical decision, the regularity effect was initially strong for low frequency words but became null after two presentations; in naming it was also initially strong but was merely reduced (although still substantial) after three repetitions. Mixed case words were recognized and named more slowly and tended to show stronger regularity effects. The results were consistent with the primary hypothesis that familiar word forms are read faster because they are processed at a larger grain-size, which requires fewer operations to achieve lexical selection. Results are discussed in terms of a neurobiological model of word recognition based on brain imaging studies.

• 전자공학회논문지C
• /
• 제35C권2호
• /
• pp.65-70
• /
• 1998

In this paper, we will to check and to correct mis-recognized word using Eojeol information. First, we divided into 16 classes that constituents in a Eojeol after we analyzed Korean statement into Eojeol units. Eojeol-Constituent state diagram constructed these constitutents, find the Left-Right Connectivity Information. As analogized the speech of connectivity information, reduced the number of cadidate words and restricted case of morphological analysis for mis-recognition Eojeol. Then, we improved correction speed uisng heuristic information as the adjacency information for Eojeol each other. In the correction phase, construct Reverse-Order Word Dictionary. Using this, we can trace word dictionary regardless of mis-recongnition word position. Its results show that improvement of recognition rate from 97.03% to 98.02% and check rate, reduction of chadidata words and morpholgical analysis cases.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 1988년도 전기.전자공학 학술대회 논문집
• /
• pp.10-13
• /
• 1988

An isolated-word recognition method using adaptively partitioned multisection codebooks is proposed. Each training utterance was divided into several sections according to its pattern extracted by labeling technique. For each pattern, reference codebooks were generated by clustering the training vectors of the same section. In recognition procedure, input speech was divided into the sections by the same method used in codebook generation procedure, and recognized to the reference word whose codebook represented the smallest average distortion. The proposed method was tested for 100 Korean words and attained recognition rate about 96 percent.

• 전자공학회논문지B
• /
• 제31B권6호
• /
• pp.17-24
• /
• 1994

This paper is a study on Korean word recognition and suggest the method that stabilizes the state-transition in the HMM by applying the `inertia' to the feature vector sequences. In order to reduce the quantized distortion considering probability distribution of input vectors, we used SOFM, an unsupervised learning method, as a vector quantizer, By applying inertia to the feature vector sequences, the overlapping of probability distributions for the response path of each word on the self organizing feature map can be reduced and the state-transition in the Hmm can be Stabilized. In order to evaluate the performance of the method, we carried out experiments for 50 DDD area names. The results showed that applying inertia to the feature vector sequence improved the recognition rate by 7.4% and can make more HMMs available without reducing the recognition rate for the SOFM having the fixed number of neuron.