• 한국통계학회:학술대회논문집
• /
• 한국통계학회 2001년도 추계학술발표회 논문집
• /
• pp.75-76
• /
• 2001

An endpoint detection method for speech signals utilizing hidden Markov model(HMM) is proposed. It turns out that the proposed algorithm is quite satisfactory to apply isolated word speech recognition.

• 음성과학
• /
• 제14권1호
• /
• pp.37-62
• /
• 2007

The current study examines the role of post-lexical tonal patterns of a prosodic phrase in word segmentation. In a word spotting experiment, native Korean listeners were asked to spot a disyllabic or trisyllabic word from twelve syllable speech stream that was composed of three Accentual Phrases (AP). Words occurred with various post-lexical intonation patterns. The results showed that listeners spotted more words in phrase-initial than in phrase-medial position, suggesting that the AP-final H tone from the preceding AP helped listeners to segment the phrase-initial word in the target AP. Results also showed that listeners' error rates were significantly lower when words occurred with initial rising tonal pattern, which is the most frequent intonational pattern imposed upon multisyllabic words in Korean, than with non-rising patterns. This result was observed both in AP-initial and in AP-medial positions, regardless of the frequency and legality of overall AP tonal patterns. Tonal cues other than initial rising tone did not positively influence the error rate. These results not only indicate that rising tone in AP-initial and AP_final position is a reliable cue for word boundary detection for Korean listeners, but further suggest that phrasal intonation contours serve as a possible word boundary cue in languages without lexical prominence.

• 정보와 통신
• /
• 제10권10호
• /
• pp.22-29
• /
• 1993

This paper describes development of a real-time voice recognition dialing system which can recognize around one hundred word vocabularies in speaker independent mode. The voice recognition algorithm is implemented on a DSP board with a telephone interface plugged in an IBM PC AT/486. In the DSP board, procedures for feature extraction, vector quantization(VQ), and end-point detection are performed simultaneously in every 10msec frame interval to satisfy real-time constraints after the word starting point detection. In addition, we optimize the VQ codebook size and the end-point detection procedure to reduce recognition time and memory requirement. The demonstration system is being displayed in MOBILAB of Korea Mobile Telecom at the Taejon EXPO '93.

• 음성과학
• /
• 제13권1호
• /
• pp.29-41
• /
• 2006

This study investigates the degree to which various prosodic cues at the boundaries of prosodic phrases in Korean contribute to word segmentation. Since most phonological words in Korean are produced as one Accentual Phrase (AP), it was hypothesized that the detection of acoustic cues at AP boundaries would facilitate word segmentation. The prosodic characteristics of Korean APs include initial strengthening at the beginning of the phrase and pitch rise and final lengthening at the end. A perception experiment utilizing an artificial language learning paradigm revealed that cues conforming to the aforementioned prosodic characteristics of Korean facilitated listeners' word segmentation. Results also indicated that duration and amplitude cues were more helpful in segmentation than pitch. Nevertheless, results did show that a pitch cue that did not conform to the Korean AP interfered with segmentation.

• Journal of information and communication convergence engineering
• /
• 제17권3호
• /
• pp.191-197
• /
• 2019

Morphological analysis is used for searching sentences and understanding context. As most morpheme analysis methods are based on predefined dictionaries, the problem of a target word not being registered in the given morpheme dictionary, the so-called unregistered word problem, can be a major cause of reduced performance. The current practical solution of such unregistered word problem is to add them by hand-write into the given dictionary. This method is a limitation that restricts the scalability and expandability of dictionaries. In order to overcome this limitation, we propose a novel method to automatically expand a dictionary by means of use-case analysis, which checks the validity of the unregistered word by exploring the use-cases through web crawling. The results show that the proposed method is a feasible one in terms of the accuracy of the validation process, the expandability of the dictionary and, after registration, the fast extraction time of morphemes.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제31권9호
• /
• pp.1171-1179
• /
• 2004

본 논문에서는 음성인식의 전처리 단계로서 음성 영역과 비음성 영역 사이의 경계를 검출하는 음성경계 추출에 대하여 기술한다. 본 논문에서는 음성경계 추출을 위해 두 가지의 특징벡터를 사용한다. 첫 번째는 백색잡음(white noise)에 강건한 시간 영역의 정보인 정규화된 RMS이고, 두 번째는 주파수 영역의 정보인 정규화된 멜주파수 대역 최대 에너지(met-frequency band maximum energy)이다. 본 논문에서 사용하는 음성경계 추출 알고리즘은 학습을 통해 규칙을 생성하고 음성의 시간 정보를 적용하기 위해 순환노드를 추가한 순환 퍼지연상기억장치이다. 퍼지부의 가중치 학습은 헤비안 학습 방법을 사용하고, 순환부의 가중치 학습을 위해서는 오류 역전파(error back-propagation) 알고리즘을 사용한다. 실험에서는 KAIST에서 제공한 연령과 성별로 구분된 음성 자료를 사용하였다.

• 말소리와 음성과학
• /
• 제4권3호
• /
• pp.25-30
• /
• 2012

This study investigated whether native Korean listeners could use visual cues for the identification of the English consonants /b/ and /v/. Both auditory and audiovisual tokens of word minimal pairs in which the target phonemes were located in word-initial or word-medial position were used. Participants were instructed to decide which consonant they heard in $2{\times}2$ conditions: cue (audio-only, audiovisual) and location (word-initial, word-medial). Mean identification scores were significantly higher for audiovisual than audio-only condition and for word-initial than word-medial condition. Also, according to signal detection theory, sensitivity, d', and response bias, c were calculated based on both hit rates and false alarm rates. The measures showed that the higher identification rate in the audiovisual condition was related with an increase in sensitivity. There were no significant differences in response bias measures across conditions. This result suggests that native Korean speakers can use visual cues while identifying confusing non-native phonemic contrasts. Visual cues can enhance non-native speech perception.

• 대한음성학회지:말소리
• /
• 제59호
• /
• pp.53-68
• /
• 2006

We report the evaluation results of the Korean speech recognition platform called ECHOS. The platform has an object-oriented and reusable architecture so that researchers can easily evaluate their own algorithms. The platform has all intrinsic modules to build a large vocabulary speech recognizer: Noise reduction, end-point detection, feature extraction, hidden Markov model (HMM)-based acoustic modeling, cross-word modeling, n-gram language modeling, n-best search, word graph generation, and Korean-specific language processing. The platform supports both lexical search trees and finite-state networks. It performs word-dependent n-best search with bigram in the forward search stage, and rescores the lattice with trigram in the backward stage. In an 8000-word continuous speech recognition task, the platform with a lexical tree increases 40% of word errors but decreases 50% of recognition time compared to the HTK platform with flat lexicon. ECHOS reduces 40% of recognition errors through incorporation of cross-word modeling. With the number of Gaussian mixtures increasing to 16, it yields word accuracy comparable to the previous lexical tree-based platform, Julius.

• 한국산업정보학회논문지
• /
• 제22권2호
• /
• pp.43-52
• /
• 2017

패턴 분석은 신호 및 영상 처리와 텍스트 마이닝 분야에서 가장 중요한 기술 중 하나이다. 이산 푸리에 변환(Discrete Fourier Transform: DFT)은 일반적으로 신호와 영상의 패턴을 분석하는데 사용된다. 본 논문에서는 DFT가 텍스트 패턴 분석에도 적용될 수 있음을 가정하고 문서의 텍스트 패턴이 다른 문서에서도 존재하는지를 탐색하는 표절 문장 탐색에 세계 최초로 적용하였다. 이를 위해 텍스트를 ASCII 코드로 변환하여 신호화하고 복사/붙여넣기, 용어의 재배치 등 단순한 표절 형태의 탐색은 Cross-Correlation(상호상관)을 이용하였다. 또한 유의어를 사용하거나 번역 및 요약 등의 표절 형태를 탐색하기 위해 워드넷(WordNet) 유사도를 사용하였다. 실험을 위해 표절 탐색 분야의 저명한 워크숍인 PAN에서 제공하는 공식적인 데이터 셋(2013 Corpus)을 사용하였으며, 실험 결과 11개의 표절 문장 탐색 기법 중 4번째로 우수한 성능을 보였다.

• 한국음향학회지
• /
• 제39권5호
• /
• pp.454-460
• /
• 2020

음성인식 기능을 제공하는 인공지능 비서들은 정확도가 뛰어난 클라우드 기반의 음성인식을 통해 동작한다. 클라우드 기반의 음성인식에서 시작 단어 인식은 대기 중인 기기를 활성화하는 데 중요한 역할을 한다. 본 논문에서는 공개 데이터셋인 구글의 Speech Commands 데이터셋을 사용하여 스펙트로그램 및 멜-주파수 캡스트럼 계수 특징을 입력으로 하여 모바일 기기에 대응한 저 연산 시작 단어 검출을 위한 합성곱 신경망의 성능을 비교한다. 본 논문에서 사용한 합성곱 신경망은 다층 퍼셉트론, 일반적인 합성곱 신경망, VGG16, VGG19, ResNet50, ResNet101, ResNet152, MobileNet이며, MobileNet의 성능을 유지하면서 모델 크기를 1/25로 줄인 네트워크도 제안한다.