The aim of this paper is to analyze vowels in voice imitation and disguised voice, and to find the invariable phonetic features of the speaker. In this paper we examined the formants of monophthongs /a, u, i, o, {$\omega},{\;}{\varepsilon},{\;}{\Lambda}$/. The results of the present are as follows : $\circled1$ Speakers change their vocal tract features. $\circled2$ Vowels /a, ${\varepsilon}$, i/ appear to be proper for speaker recognition since they show invariable acoustic feature during voice modulation. $\circled3$ F1 does not change easily compared to higher formants. $\circled4$ F3-F2 appears to be constituent for a speaker identification in vowel /a/ and /$\varepsilon$/, and F4-F2 in vowel /i/. $\circled5$ Resulting of F-ratio, differences of each formants were more useful than individual formant of a vowel to speaker recognition.

The purpose of this dissertation is to make the general korean speakers to learn the standard pronunciations. But it is in existence that the obstructions of the command of the standard pronunciations. They are the mistake in the education course on the korean pronunciations, the teacher's capability and the mass communications's duplicity. To overcome this obstructions, we must concentrate our efforts on the propagation of the standard pronunciations. To propagate of the standard pronunciations we can take a several method. These are the presentation of the pronunciation mistakes, audio-visual teaching, the presentation of the pronunciation principles and the use of the korean dictionary. The standard pronunciations are different from the pronunciations of the general korean speakers in many respects. So we can't make an accurate estimate of the pronunciation's changes. No one knows what will happen in the future about the korean pronunciations. But we must teach the standard pronunciations to the general korean speakers. The standard pronunciations are offically valid in the present time.

Working memory is the term used to refer to the mental workplace in which information can be temporarily stored and manipulated during complex everyday activities such as understanding language. The studies on language and working memory are based on Baddeley's phonological working memory and Daneman and Carpenter's functional working memory. This article reviews two working memory models and the studies on language and working memory based on each model. These are described in the implication of working memory in language development and specific language impairment-evaluation and treatment.

Comprehension monitoring is a process of message evaluation and a very important skill of communication. Comprehension monitoring is a necessary language skill in classroom, as it is important for children to assess their own understanding of task instructions and teaching content. The present study investigated comprehension monitoring skill of children with Specific Language Impairment(SLI) compared to age-matched children and language-matched children. 18 vignettes and 6 displays were constructed. Children were asked to choose one object or 'DK' card from a display after the children listened to the vignettes. The results showed that children with SLI didn't have problem in comprehending unambiguous utterance and using prior statement. However, they had problem in monitoring comprehending ambiguous utterance compared age-matched children. Their performance were similar to language matched younger children. With respect to their performance, several possible explanations were discussed.

In this paper, we propose the automatic speech database verification method(or called automatic verification) based on confidence measure for a large speech database. This method verifies the consistency between given transcription and speech using the confidence measure. The automatic verification process consists of two stages : the word-level likelihood computation stage and multi-level likelihood ratio computation stage. In the word-level likelihood computation stage, we calculate the word-level likelihood using the viterbi decoding algorithm and make the segment information. In the multi-level likelihood ratio computation stage, we calculate the word-level and the phone-level likelihood ratio based on confidence measure with anti-phone model. By automatic verification, we have achieved about 61% error reduction. And also we can reduce the verification time from 1 month in manual to 1-2 days in automatic.

It is almost impossible to generate the sound field effect in real time with the time-domain linear convolution because of its large multiplication operation requirement. To solve this, three methods are introduced to reduce the number of multiplication operations in this paper. Firstly, the time-domain linear convolution is replaced with the frequency-domain circular convolution. In other words, the linear convolution result can be derived from that of the circular convolution. This technique reduces the number of multiplication operations remarkably, Secondly, a subframe concept is introduced, i.e., one original frame is divided into several subframes. Then the FFT is executed for each subframe and, as a result, the number of multiplication operations can be reduced. Finally, the QFT is used in stead of the FFT. By combining all the above three methods into our final the SFE generation algorithm, the number of computations are reduced sufficiently and the real-time SFE generation becomes possible with a general PC.

In this parer, we propose LM adaptation for broadcast news recognition. We collect information of recent articles from the internet on real time, make a recent small size LM, and then interpolate recent LM with a existing LM composed of existing large broadcast news corpus. We performed interpolation experiments to get the best type of articles from recent corpus because collected recent corpus is composed of articles which are related with test set, and which are unrelated. When we made an adapted LM using recent LM with similar articles to test set through Tf-Idf method and existing LM, we got the best result that ERR of pseudo-morpheme based recognition performance has 17.2 % improvement and the number of OOV has reduction from 70 to 27.

In this paper, we introduce the multimodal user interface technology, especially based on speech. We classify multimodal interface technologies into four classes: sequential, alternate, supplementary, and semantic multimodal interfaces. After introducing four types of multimodal interfaces, we explain standard activities currently being activated.

A weather forecast service with speech recognition is described. This service allows users to get the weather information of all the cities by saying the city names with just one phone call, which was not provided in the previous weather forecast service. Speech recognition is implemented in the intelligent peripheral (IP) of the advanced intelligent network (AIN). The AIN is a telephone network architecture that separates service logic from switching equipment, allowing new services to be added without having to redesign switches to support new services. Experiments in speech recognition show that the recognition accuracy is 90.06% for the general users' speech database. For the laboratory members' speech database, the accuracies are 95.04% and 93.81%, respectively in simulation and in the test on the developed system.

For educational and research purposes, a Korean speech recognition platform is designed. It is based on an object-oriented architecture and can be easily modified so that researchers can readily evaluate the performance of a recognition algorithm of interest. This platform will save development time for many who are interested in speech recognition. The platform includes the following modules: Noise reduction, end-point detection, met-frequency cepstral coefficient (MFCC) and perceptually linear prediction (PLP)-based feature extraction, hidden Markov model (HMM)-based acoustic modeling, n-gram language modeling, n-best search, and Korean language processing. The decoder of the platform can handle both lexical search trees for large vocabulary speech recognition and finite-state networks for small-to-medium vocabulary speech recognition. It performs word-dependent n-best search algorithm with a bigram language model in the first forward search stage and then extracts a word lattice and restores each lattice path with a trigram language model in the second stage.