• The Journal of the Acoustical Society of Korea
• /
• v.39 no.2
• /
• pp.131-136
• /
• 2020

Wake Up Word (WUW) is a short utterance used to convert speech recognizer to recognition mode. The WUW defined by the user who actually use the speech recognizer is called user-defined WUW. In this paper, to recognize user-defined WUW, we construct traditional Gaussian Mixture Model-Hidden Markov Model (GMM-HMM), Linear Discriminant Analysis (LDA)-GMM-HMM and LDA-Deep Neural Network (DNN)-HMM based system and compare their performances. Also, to improve recognition accuracy of the WUW system, a threshold method is applied to each model, which significantly reduces the error rate of the WUW recognition and the rejection failure rate of non-WUW simultaneously. For LDA-DNN-HMM system, when the WUW error rate is 9.84 %, the rejection failure rate of non-WUW is 0.0058 %, which is about 4.82 times lower than the LDA-GMM-HMM system. These results demonstrate that LDA-DNN-HMM model developed in this paper proves to be highly effective for constructing user-defined WUW recognition system.

• Clinical and Experimental Pediatrics
• /
• v.64 no.10
• /
• pp.497-503
• /
• 2021

Reading disability (dyslexia) refers to an unexpected difficulty with reading for an individual who has the intelligence to be a much better reader. Dyslexia is most commonly caused by a difficulty in phonological processing (the appreciation of the individual sounds of spoken language), which affects the ability of an individual to speak, read, and spell. In this paper, I describe reading disabilities by focusing on their underlying neurobiological mechanisms. Neurobiological studies using functional brain imaging have uncovered the reading pathways, brain regions involved in reading, and neurobiological abnormalities of dyslexia. The reading pathway is in the order of visual analysis, letter recognition, word recognition, meaning (semantics), phonological processing, and speech production. According to functional neuroimaging studies, the important areas of the brain related to reading include the inferior frontal cortex (Broca's area), the midtemporal lobe region, the inferior parieto-temporal area, and the left occipitotemporal region (visual word form area). Interventions for dyslexia can affect reading ability by causing changes in brain function and structure. An accurate diagnosis and timely specialized intervention are important in children with dyslexia. In cases in which national infant development screening tests have been conducted, as in Korea, if language developmental delay and early predictors of dyslexia are detected, careful observation of the progression to dyslexia and early intervention should be made.

• Proceedings of the Korean Society for Cognitive Science Conference
• /
• 2005.05a
• /
• pp.141-144
• /
• 2005

본 연구에서는 한국어 음성어휘의 표상이 음성정보로 이루어져 있는지 아니면 철자정보로 이루어져 있는지와, 음성어휘의 표상단위가 음절인지를 조사하였다. 연구 과제로는 형태점화기법(form priming technique)을 사용하였다. 점화 자극과 목표 자극간의 관련성 수준으로는 음운적 혹은 철자적인 것을 사용하였다. 점화 자극과 목표자극간의 간격은 50msec 이었다. 따라서 SOA 각 점화자극의 제시시간에 50msec를 합한 것이다. 연구 결과는 음운적으로 관련되어 있을 때에는 무관련 조건에 비해 느려지는 방해점화(inhibitory priming)가 나타났고 철자적으로 관련되어 있는 경우에는 촉진적점화(facilitatory priming)가 나타났다. 이런 연구 결과는 시각단어 재인 연구에서도 보이는 공통적인 현상으로 음성어휘와 시각어휘가 심성어휘표상에서는 동일한 형태를 지니고 있음을 암시한다. 또한 음운적 관련성에 의한 방해점화효과는 음운음절을 공유하고 있는 어휘 후보자간의 경쟁 때문에 나타난 것으로 해석되며, 철자관련 촉진효과는 어휘접근 전에 일어나는 과정에서 나타나는 것으로 해석된다.

• MALSORI
• /
• no.65
• /
• pp.93-103
• /
• 2008

In this paper, we propose a method for optimizing a multiple pronunciation dictionary used for modeling pronunciation variations of non-native speech. The proposed method removes some confusable pronunciation variants in the dictionary, resulting in a reduced dictionary size and less decoding time for automatic speech recognition (ASR). To this end, a confusability measure is first defined based on the Levenshtein distance between two different pronunciation variants. Then, the number of phonemes for each pronunciation variant is incorporated into the confusability measure to compensate for ASR errors due to words of a shorter length. We investigate the effect of the proposed method on ASR performance, where Korean is selected as the target language and Korean utterances spoken by Chinese native speakers are considered as non-native speech. It is shown from the experiments that an ASR system using the multiple pronunciation dictionary optimized by the proposed method can provide a relative average word error rate reduction of 6.25%, with 11.67% less ASR decoding time, as compared with that using a multiple pronunciation dictionary without the optimization.

• Proceedings of the IEEK Conference
• /
• 2007.07a
• /
• pp.303-304
• /
• 2007

In this paper, we implement an automatic distance speech recognition system for voiced-enabled services. We first construct a baseline automatic speech recognition (ASR) system, where acoustic models are trained from speech utterances spoken by using a cross-talking microphone. In order to improve the performance of the baseline ASR using distance speech, the acoustic models are adapted to adjust the spectral characteristics of speech according to different microphones and the environmental mismatches between cross-talking and distance speech. Next we develop a voice activity detection algorithm for distance speech. We compare the performance of the base-line system and the developed ASR system on a task of PBW (Phonetically Balanced Word) 452. As a result it is shown that the developed ASR system provides the average word error rate (WER) reduction of 30.6 % compared to the baseline ASR system.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.3
• /
• pp.357-362
• /
• 2004

Language models are essential in predicting the next word in a spoken sentence, thereby enhancing the speech recognition accuracy, among other things. However, spoken language domains are too numerous, and therefore developers suffer from the lack of corpora with sufficient sizes. This paper proposes a method of combining two n-gram language models, one constructed from a very small corpus of the right domain of interest, the other constructed from a large but less adequate corpus, resulting in a significantly enhanced language model. This method is based on the observation that a small corpus from the right domain has high quality n-grams but has serious sparseness problem, while a large corpus from a different domain has more n-gram statistics but incorrectly biased. With our approach, two n-gram statistics are combined by extending the idea of Katz's backoff and therefore is called a dual-source backoff. We ran experiments with 3-gram language models constructed from newspaper corpora of several million to tens of million words together with models from smaller broadcast news corpora. The target domain was broadcast news. We obtained significant improvement (30%) by incorporating a small corpus around one thirtieth size of the newspaper corpus.

• Speech Sciences
• /
• v.14 no.4
• /
• pp.7-24
• /
• 2007

This paper discusses some current issues regarding how prosodic structure is manifested in fine-grained phonetic details, how prosodically-conditioned articulatory variation is explained in terms of speech dynamics, and how such phonetic manifestation of prosodic structure may be exploited in spoken word recognition. Prosodic structure is phonetically manifested in prosodically important landmark locations such as prosodic domain-final position, domain-initial position and stressed/accented syllables. It will be discussed how each of the prosodic landmarks engenders particular phonetic patterns, ow articulatory variation in such locations are dynamically accounted for, and how prosodically-driven fine-grained phonetic detail is exploited by listeners in speech comprehension.

• Phonetics and Speech Sciences
• /
• v.2 no.3
• /
• pp.115-124
• /
• 2010

The current study investigated the role of acoustic correlates of domain-initial strengthening in lexical segmentation of a non-native language. In a series of cross-modal identity-priming experiments, native Korean listeners heard English auditory stimuli and made lexical decision to visual targets (i.e., written words). The auditory stimuli contained critical two word sequences which created temporal lexical ambiguity (e.g., 'mill#company', with the competitor 'milk'). There was either an IP boundary or a word boundary between the two words in the critical sequences. The initial CV of the second word (e.g., [$k_{\Lambda}$] in 'company') was spliced from another token of the sequence in IP- or Wd-initial positions. The prime words were postboundary words (e.g., company) in Experiment 1, and preboundary words (e.g., mill) in Experiment 2. In both experiments, Korean listeners showed priming effects only in IP contexts, indicating that they can make use of IP boundary cues of English in lexical segmentation of English. The acoustic correlates of domain-initial strengthening were also exploited by Korean listeners, but significant effects were found only for the segmentation of postboundary words. The results therefore indicate that L2 listeners can make use of prosodically driven phonetic detail in lexical segmentation of L2, as long as the direction of those cues are similar in their L1 and L2. The exact use of the cues by Korean listeners was, however, different from that found with native English listeners in Cho, McQueen, and Cox (2007). The differential use of the prosodically driven phonetic cues by the native and non-native listeners are thus discussed.

• 전자공학회논문지 IE
• /
• v.45 no.3
• /
• pp.20-26
• /
• 2008

Two independent component analysis(ICA) algorithms were applied for the recognition of speech signals corrupted by a car engine noise. Speech recognition was performed by hidden markov model(HMM) for the estimated signals and recognition rates were compared with those of orginal speech signals which are not corrupted. Two different ICA methods were applied for the estimation of speech signals, one of which is FastICA algorithm that maximizes negentropy, the other is information-maximization approach that maximizes the mutual information between inputs and outputs to give maximum independence among outputs. Word recognition rate for the Korean news sentences spoken by a male anchor is 87.85%, while there is 1.65% drop of performance on the average for the estimated speech signals by FastICA and 2.02% by information-maximization for the various signal to noise ratio(SNR). There is little difference between the methods.

• 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%.