Locus equation scatterplots for [bilabial stop + vowel] syllables were obtained from 16 non-native speakers of English and Russian. The results indicated that both Russian speakers of English and English speakers of Russian exhibited modifications towards respective L2 norms in slopes and y-intercepts. All non-native locus equations generated exhibited linearity. Accordingly, the basic results reported in [17] were reverified by securing a larger subject base. More experienced speakers displayed better approximations to L2 norms than less experienced speakers, indicating the necessity of perception- and articulation-related learning for allophonic variations due to adjacent phonetic environments.

The purpose of this paper is to compare linguistic features of meter in Amoretti and Sonnets from the Portuguese. Amoretti was composed in the 16th century and Sonnets from the Portuguese in the 19th century. Two poems are very strict in satisfying $^*<_I$(LS that makes a lexical stress matched on a strong position. But they allow a peaked monosyllabic word to be matched on a weak position only if it does not violate $^*(\dot{\sigma}\;{\sigma})$. Syntactic inversions which are found in two poems occur to satisfy metrical constraints and rhyme. Particularly, syntactic inversions for rhyme are mainly found in Amoretti, but are rare in Sonnets from the Portuguese which includes many enjambment lines. Furthermore, this paper shows that syntactic inversions can be explained by the interaction of constraints, such as metrical constraints, SCI, and Binary Foot.

This study aims at exploring the degree of phonetic similarity between Korean and Japanese vowels in terms of acoustic features by performing the speech production test on Korean speakers and Japanese speakers. For this purpose, the speech of 16 Japanese speakers for Japanese speech data, and the speech of 16 Korean speakers for Korean speech data were utilized. The findings in assessing the degree of the similarity of the 7 nearest equivalents of the Korean and Japanese vowels are as follows: First, Korean /i/ and /e/ turned out to display no significant differences in terms of F1 and F2 with their counterparts, Japanese /i/ and /e/, and the distribution of F1 and F2 of Korean /i/ and /e/ in the distributional map completely overlapped with Japanese /i/ and /e/. Accordingly, Korean /i/ and /e/ were believed to be "identical." Second, Korean /a/, /o/, and /i/ displayed a significant difference in either F1 or F2, but showed a great similarity in distribution of F1 and F2 with Japanese /a/, /o/, and /m/ respectively. Korean /a/ /o/, and /i/, therefore, were categorized as very similar to Japanese vowels. Third, Korean /u/, which has the counterpart /m/ in Japanese, showed a significant difference in both F1 and F2, and only half of the distribution overlapped. Thus, Korean /u/ was analyzed as being a moderately similar vowel to Japanese vowels. Fourth, Korean /${\wedge}$/ did not have a close counterpart in Japanese, and was classified as "the least similar vowel."

Verb phrase could have two different meanings according to which is followed after verb; adverb or preposition. The meaning of 'verb+adverb' is deduced from a figurative meaning which is idiomatic expression, and 'verb+preposition' is interpreted as the literal meaning. The purpose of this study is to observe how English native speakers and Korean leaners of English distinguish two sentences of the same word strings with acoustic cues like pause and duration. According to the result, as pause was used for meaning distinction, it was likely that the pause length preceding prepositions was longer than that of following adverbs. To distinguish two sentences of the same word strings, all participants seemed to use pause, verb lengthening and adverb/preposition lengthening. Among them, there is a hierarchical significance; in sequence, pause, verb lengthening, adverb/preposition lengthening.

This paper aims to propose how to extract intonation patterns using Momel, a pitch stylization algorithm, and to present results of analyzing speech corpora in comparison with those in earlier researches. Two speech corpora are used: one is the sound files obtained from the K-ToBI web site, and the other consists of 80 passages pronounced by 4 speakers (2 male and 2 female). The results show that Momel provides significant pitch targets which can be labeled as H and L tones within prosodic units such as Accentual Phrase (AP) and Intonation Phrase (IP). The resulting AP patterns and IP boundary tone patterns correspond to those in earlier researches. Thus, this study will contribute to the study of intonation as well as to the development of automatic intonation labeling systems.

Spoken dialog system development includes many laborious and inefficient tasks. Since there are many components such as speech recognition, language understanding, dialog management and knowledge management in a spoken dialog system, a developer should take an effort to edit corpus and train each model separately. To reduce a cost for editing corpus and training each model, we need more systematic and efficient working environment. For the working environment, we propose DialogStudio as a spoken dialog system workbench.

In this paper, we propose a use of confidence vector as an intermediate input feature for multi-stage based speech recognition architecture to improve recognition accuracy. A multi-stage speech recognition structure is introduced as a method to reduce the computational complexity of the decoding procedure and then accomplish faster speech recognition. Conventional multi-stage speech recognition is usually composed of three stages, acoustic search, lexical search, and acoustic re-scoring. In this paper, we focus on improving the accuracy of the lexical decoding by introducing a confidence vector as an input feature instead of phoneme which was used typically. We take experimental results on 220K Korean Point-of-Interest (POI) domain and the experimental results show that the proposed method contributes on improving accuracy.

Recently, many techniques have been proposed to improve the noise robustness for speaker verification. In this paper, we consider the feature recombination technique in multi-band approach. In the conventional feature recombination for speaker verification, to compute the likelihoods of speaker models or universal background model, whole feature components are used. This computation method is not effective in a view point of multi-band approach. To deal with non-effectiveness of the conventional feature recombination technique, we introduce a subband likelihood computation, and propose a modified feature recombination using subband likelihoods. In decision step of speaker verification system in noise environments, a few very low likelihood scores of a speaker model or universal background model cause speaker verification system to make wrong decision. To overcome this problem, a reliable feature selection method is proposed. The low likelihood scores of unreliable feature are substituted by likelihood scores of the adaptive noise model. In here, this adaptive noise model is estimated by maximum a posteriori adaptation technique using noise features directly obtained from noisy test speech. The proposed method using subband-based reliable feature selection obtains better performance than conventional feature recombination system. The error reduction rate is more than 31 % compared with the feature recombination-based speaker verification system.

As the Internet becomes prevalent in our lives, harmful contents, such as phornographic videos, have been increasing on the Internet, which has become a very serious problem. To prevent such an event, there are many filtering systems mainly based on the keyword-or image-based methods. The main purpose of this paper is to devise a system that classifies pornographic videos based on the audio information. We use the mel-cepstrum modulation energy (MCME) which is a modulation energy calculated on the time trajectory of the mel-frequency cepstral coefficients (MFCC) as well as the MFCC as the feature vector. For the classifier, we use the well-known Gaussian mixture model (GMM). The experimental results showed that the proposed system effectively classified 98.3% of pornographic data and 99.8% of non-pornographic data. We expect the proposed method can be applied to the more accurate classification system which uses both video and audio information.

It is well known that when there is an acoustic mismatch between the speech obtained during training and testing, the accuracy of speaker verification systems drastically deteriorates. This paper presents the use of MFCCs' histogram enhancement technique in order to improve the robustness of a speaker verification system. The technique transforms the features extracted from speech within an utterance such that their statistics conform to reference distributions. The reference distributions proposed in this paper are uniform distribution and beta distribution. The transformation modifies the contrast of MFCCs' histogram so that the performance of a speaker verification system is improved both in the clean training and testing environment and in the clean training and noisy testing environment.