• Proceedings of the KSPS conference
• /
• 2000.07a
• /
• pp.312-320
• /
• 2000

This presentation aims to explore some possibility of non-segmental phonetics usually ignored in phonetics education. In pedagogical phonetics, especially ESL/EFL oriented phonetics speech sounds tend to be classified in two criteria 1) 'pronunciation' which deals with segments and 2) 'prosody' or 'suprasegmentals', a criterion that deals with non-segmental elements such as stress and intonation. However, speech involves more dynamic processing. It is non-linear and multi-dimensional in spite of the linear sequence of symbols in phonetic/phonological transcriptions. No word is without pitch or voice quality apart from segmental characteristics whether it is spoken in isolation or cut out from continuous speech. This simply tells the dichotomy of pronunciation and prosody is merely a useful convention. There exists some room to consider dynamic non-segmental phonetics. Examples of non-segmental phonetic investigation, some of the analyses conducted within the frame of Firthian Prosodic Analysis, especially of the relation between vowel variants and foot types, are examined and we see what kind of auditory phonetic training is required to understand impressionistic transcriptions which lie behind the non-segmental phonetics.

• Proceedings of the IEEK Conference
• /
• 2000.07a
• /
• pp.257-259
• /
• 2000

In this paper, a word-unit recognition is performed to implement a speech recognition system over the web, using JAVA Applet and continuous distributed HMM. The system based on Client/server model is designed. A client computer processes speech with Applet, and then transmits feature parameters to the server computer though the Internet. The speech recognition system in the server computer transmits the result applied by the forward algorithm to the client computer and the result is displayed in the client computer by text.

• Speech Sciences
• /
• v.11 no.4
• /
• pp.75-88
• /
• 2004

In this paper, we propose a new approach to sequential linear regression adaptation of continuous density hidden Markov models (CDHMMs) based on transformation space model (TSM). The proposed TSM which characterizes the a priori knowledge of the training speakers associated with maximum likelihood linear regression (MLLR) matrix parameters is effectively described in terms of the latent variable models. The TSM provides various sources of information such as the correlation information, the prior distribution, and the prior knowledge of the regression parameters that are very useful for rapid adaptation. The quasi-Bayes (QB) estimation algorithm is formulated to incrementally update the hyperparameters of the TSM and regression matrices simultaneously. Experimental results showed that the proposed TSM approach is better than that of the conventional quasi-Bayes linear regression (QBLR) algorithm for a small amount of adaptation data.

• Speech Sciences
• /
• v.10 no.1
• /
• pp.71-84
• /
• 2003

In the acoustic modeling for large vocabulary speech recognition, a sparse data problem caused by a huge number of context-dependent (CD) models usually leads the estimated models to being unreliable. In this paper, we develop a new clustering method based on the C45 decision-tree learning algorithm that effectively encapsulates the CD modeling. The proposed scheme essentially constructs a supervised decision rule and applies over the pre-clustered triphones using the C45 algorithm, which is known to effectively search through the attributes of the training instances and extract the attribute that best separates the given examples. In particular, the data driven method is used as a clustering algorithm while its result is used as the learning target of the C45 algorithm. This scheme has been shown to be effective particularly over the database of low unknown-context ratio in terms of recognition performance. For speaker-independent, task-independent continuous speech recognition task, the proposed method reduced the percent accuracy WER by 3.93% compared to the existing rule-based methods.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1994.06c
• /
• pp.358-363
• /
• 1994

This paper presents the improvement of the SCHMM using discrete VQ and One-Pass algorithm for keywords recognition in Korean continuous speech. The SCHMM using discrete VQ is a simple model that is composed of a variable mixture gaussian probability density function with dynamic mixture number. One-Pass algorithm is improved such that recognition rates are enhanced by fathoming any undesirable semisyllable with the low likelihood and the high duration penalty, and computation time is reduced by testing only the frame which is dissimilar to the previously testd frame. In recognition experiments for speaker-dependent case, the improved One-Pass algorithm has shown recognition rates as high as 99.7% and has reduced compution time by about 30% compared with the currently abailable one-pass algorithm.

• 제어로봇시스템학회:학술대회논문집
• /
• 1993.10a
• /
• pp.870-876
• /
• 1993

This paper presents a binary clustering network (BCN) and a heuristic algorithm to detect pitch for recognition of keywords in continuous speech. In order to classify nonlinear patterns, BCN separates patterns into binary clusters hierarchically and links same patterns at root level by using the supervised learning and the unsupervised learning. BCN has many desirable properties such as flexibility of dynamic structure, high classification accuracy, short learning time, and short recall time. Pitch Detection algorithm is a heuristic model that can solve the difficulties such as scaling invariance, time warping, time-shift invariance, and redundance. This recognition algorithm has shown recognition rates as high as 95% for speaker-dependent as well as multispeaker-dependent tests.

• Speech Sciences
• /
• v.7 no.4
• /
• pp.11-26
• /
• 2000

In this paper, we propose the HMM with the MIN module. Because initial and re-estimated variance vectors are important elements for performance in HMM recognition systems, we propose a method which compensates for the mismatched statistical feature of training and test data. The MIN module function is a differentiable function similar to the sigmoid function. Unlike a continuous density function, it does not include variance vectors of the data set. The proposed hybrid HMM/MIN module is a unified network in which the observation probability in the HMM is replaced by the MIN module neural network. The parameters in the unified network are re-estimated by the gradient descent method for the Maximum Likelihood (ML) criterion. In estimating parameters, the variance vector is not estimated because there is no variance element in the MIN module function. The experiment was performed to compare the performance of the proposed HMM and the conventional HMM. The experiment measured an isolated number for speaker independent recognition.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.5
• /
• pp.44-51
• /
• 1999

This paper presents a massively parallel computational model for the efficient integration of speech and natural language understanding. The phoneme model is based on continuous Hidden Markov Model with context dependent phonemes, and the language model is based on a knowledge base approach. To construct the knowledge base, we adopt a hierarchically-structured semantic network and a memory-based parsing technique that employs parallel marker-passing as an inference mechanism. Our parallel speech recognition algorithm is implemented in a multi-Transputer system using distributed-memory MIMD multiprocessors. Experimental results show that the parallel speech recognition system performs better in recognition accuracy than a word network-based speech recognition system. The recognition accuracy is further improved by applying code-phoneme statistics. Besides, speedup experiments demonstrate the possibility of constructing a realtime parallel speech recognition system.

• 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%.

• Proceedings of the Korean Information Science Society Conference
• /
• 1998.10a
• /
• pp.747-749
• /
• 1998

본 논문에서는 음성과 자연언어의 통합처리를 위한 효과적인 병렬 계산 모델을 제안한다. 음소모델은 continuous HMM에 기반을 둔 문맥종속형 음소를 사용하며, 언어모델은 knowledge-based approach를 사용한다. 또한 계층구조의 지식베이스상에서 다수의 가설을 처리하기 위해 memory-based parsing기술을 사용하였다. 본 연구의 병렬 음성인식 알고리즘은 분산메모리 MIMD 구조의 다중 Transputer 시스템을 이용하여 구현되었다. 실험을 통하여 음성인식 과정에서 발생하는 speech-specific problem의 해를 제공하고 음성인식 시스템의 병렬화를 통하여 실시간 음성인식의 가능성을 보여준다.