• 음성과학
• /
• 제6권
• /
• pp.103-117
• /
• 1999

This paper suggests that the onset-coda asymmetry found in languages like Korean and others should be dealt with in terms of one acoustic feature rather than other articulatory features, claiming that the acoustic feature involved here is [noise], i.e., 'aperiodic waveform energy'. It determines the structural well-formedness of the languages in question whether a coda ends in [noise] or not, regardless of the intensity, the frequency, and the time duration of the [noise]. Fricatives, affricates, aspirated stops, tense stops, and released stops are all disallowed in the coda position due to the acoustic feature [noise] they, commonly end with if they were, posited in the coda. The proposal implies that the three seemingly separate prohibitions of consonants in the coda position -- i) no fricatives/affricates, ii) no aspirated/tense stops, and iii) no released stops -- are directly correlated with each other. Incorporation of the one acoustic feature [noise] in the feature theory enables us to see that the aspects of onset-coda asymmetry are derived from one single source: ban, of [noise] in the coda.

• ETRI Journal
• /
• 제30권5호
• /
• pp.753-755
• /
• 2008

In this letter, we propose a new histogram equalization technique for feature compensation in speech recognition under noisy environments. The proposed approach combines a signal-to-noise-ratio-dependent feature reconstruction method and the class histogram equalization technique to effectively reduce the acoustic mismatch present in noisy speech features. Experimental results from the Aurora 2 task confirm the superiority of the proposed approach for acoustic feature compensation.

• 로봇학회논문지
• /
• 제15권1호
• /
• pp.1-7
• /
• 2020

In this paper, we introduce the methodology that utilizes deep learning-based front-end to enhance underwater feature matching. Both optical camera and sonar are widely applicable sensors in underwater research, however, each sensor has its own weaknesses, such as light condition and turbidity for the optic camera, and noise for sonar. To overcome the problems, we proposed the opti-acoustic transformation method. Since feature detection in sonar image is challenging, we converted the sonar image to an optic style image. Maintaining the main contents in the sonar image, CNN-based style transfer method changed the style of the image that facilitates feature detection. Finally, we verified our result using cosine similarity comparison and feature matching against the original optic image.

• 한국생산제조학회지
• /
• 제6권4호
• /
• pp.46-53
• /
• 1997

This paper presents a method of detection localized defects on tapered roller bearing in main spindle of machine tool system. The feature vectors, i.e. statistical parameters, in time-domain analysis technique have been calculated to extract useful features from acoustic emission signals. These feature vectors are used as the input feature of an neural network to classify and detect bearing defects. As a results, the detection of bearing defect conditions could be sucessfully performed by using an neural network with statistical parameters of acoustic emission signals.

• 대한전기학회논문지:전기물성ㆍ응용부문C
• /
• 제50권3호
• /
• pp.121-129
• /
• 2001

In this paper, a new efficient feature extraction method based on Daubechies discrete wavelet transform is presented. This paper especially deals with the assessment of process statistical parameter using the features extracted from the wavelet coefficients of measured acoustic emission signals. Since the parameter assessment using all wavelet coefficients will often turn out leads to inefficient or inaccurate results, we selected that level-3 stage of multi decomposition in discrete wavelet transform. We make use of the feature extraction parameter namely, maximum value of acoustic emission signal, average value, dispersion, skewness, kurtosis, etc. The effectiveness of this new method has been verified on ability a diagnosis transformer go through feature extraction in stage of aging(the early period, the middle period, the last period)

• 한국산업정보학회논문지
• /
• 제6권3호
• /
• pp.107-114
• /
• 2001

본 논문에서는 한국어의 음성학적 규칙을 적용한 연속 숫자음 인식에 관하여 연구한다. 연속 숫자음의 인식률은 일반적으로 음성인식 시스템 중에서 낮은 인식률을 나타낸다. 따라서 숫자음에 대하여 강건한 모델을 작성하기 위하여 음성 특징 파라미터와 음성학적 규칙을 적용하고 실험을 통하여 그 유효성을 확인하고자 한다. 이를 위하여 음성자료로는 국어공학센터(KLE)에서 채록한 4연속 숫자음을 사용하며 인식의 기본단위로서는 음성학적 규칙을 적용한 19개의 연속분포 HMM을 유사음소 단위(PLUs)로 사용한다. 또한, 인식실험에 있어서는 일반적인 멜 켑스트럽과 회귀계수를 이용한 경우와 음성학적 규칙과 특징을 확장하여 모델을 작성한 경우에 대해서 유한상태 오토마타(Finite State Automata ; FSA)에 의한 구문제어를 통한 OPDP(One Pass Dynamic Programming) 법으로 인식실험을 수행하여 그 결과를 비교 검토하였다. 그 결과, 멜 켑스트럼만을 사용한 경우 55.4%, 멜 켑스트럼과 회귀계수를 사용한 경우에는 64.6%, 특징 파라미터를 확장한 경우 74.3%, 음성학적 특징까지 고려한 경우 75.4%로 기존의 경우보다 높은 인식률을 보였다. 따라서, 음성 특징 파라미터를 확장하고 음성학적 규칙까지 함께 적용한 경우 비교적 높은 인식률을 보여 제안된 방법이 연속 숫자음 인식에 유효함을 확인하였다.

• ETRI Journal
• /
• 제41권2호
• /
• pp.235-241
• /
• 2019

In this paper, we propose a rank-weighted reconstruction feature to improve the robustness of a feed-forward deep neural network (FFDNN)-based acoustic model. In the FFDNN-based acoustic model, an input feature is constructed by vectorizing a submatrix that is created by slicing the feature vectors of frames within a context window. In this type of feature construction, the appropriate context window size is important because it determines the amount of trivial or discriminative information, such as redundancy, or temporal context of the input features. However, we ascertained whether a single parameter is sufficiently able to control the quantity of information. Therefore, we investigated the input feature construction from the perspectives of rank and nullity, and proposed a rank-weighted reconstruction feature herein, that allows for the retention of speech information components and the reduction in trivial components. The proposed method was evaluated in the TIMIT phone recognition and Wall Street Journal (WSJ) domains. The proposed method reduced the phone error rate of the TIMIT domain from 18.4% to 18.0%, and the word error rate of the WSJ domain from 4.70% to 4.43%.

• 한국음향학회지
• /
• 제33권4호
• /
• pp.225-231
• /
• 2014

본 논문은 해양에서 발생하는 음향 표적에 대해서 표적의 식별 정보 피드백을 통해 관심 표적 추적 성능을 향상시키는 기법을 제안한다. 기존의 다중주파수 표적 추적 기법은 탐지된 음향신호 특성만 활용하여 추적을 수행하는 반면, 본 논문에서 제안하는 기법은 탐지된 음향신호 특성과 더불어 관심 표적의 특징인자가 식별되는지 여부를 추적에 활용한다. 제안하는 기법은 특징인자가 식별되는 측정치에 대해서만 추적을 수행하여 관심 표적이 아닌 표적에 대한 불필요한 추적을 방지한다. 또한 특징인자가 식별된 표적을 추적할 때 동적으로 추적 파라메터를 최적화하여 관심 표적에 대해 지속적이고 안정적인 추적을 수행한다. 모의 신호를 이용하여 실험을 수행하였으며 미약한 신호가 나타나는 수중 표적에 대해 제안하는 기법을 적용한 결과, 불필요한 표적 개수가 감소하고 관심 표적의 추적이 지속적으로 안정되게 수행됨을 확인하였다.

• The Journal of the Acoustical Society of Korea
• /
• 제22권1E호
• /
• pp.43-48
• /
• 2003

The effects of linear acoustic channels have been analyzed and compensated at mel-frequency feature domain. Unlike popular RASTA filtering our approach incorporates separate filters for each mel-frequency band, which results in better recognition performance for heavy-reverberated speeches.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권10호
• /
• pp.5078-5094
• /
• 2019

In recent years, the development of drone technologies has promoted the widespread commercial application of drones. However, the ability of drone to carry explosives and other destructive materials may bring serious threats to public safety. In order to reduce these threats from illegal drones, acoustic feature extraction and classification technologies are introduced for drone sound identification. In this paper, we introduce the acoustic feature vector extraction method of harmonic line association (HLA), and subband power feature extraction based on wavelet packet transform (WPT). We propose a feature vector extraction method based on combined HLA and WPT to extract more sophisticated characteristics of sound. Moreover, to identify drone sounds, support vector machine (SVM) classification with the optimized parameter by genetic algorithm (GA) is employed based on the extracted feature vector. Four drones' sounds and other kinds of sounds existing in outdoor environment are used to evaluate the performance of the proposed method. The experimental results show that with the proposed method, identification probability can achieve up to 100 % in trials, and robustness against noise is also significantly improved.