This study was designed to investigate the relationship between intelligibility and phonemic contrasts, and acoustic features in terms of speech development. A total of 212 Putonghua speaking children was participated in the experiment. There were phonemic contrasts significantly related with speech intelligibility: aspirated vs. fricative, retroflex vs. unretroflex, and front vs. back nasal vowel contrast. A regression analysis showed that 88% of the speech intelligibility could be predicted by these phonemic contrasts. Acoustic values were significantly related to the intelligibility of the Putonghua-speaking children's speech: voice onset time of unaspirated stops, and the duration of frication noise in fricatives.

In previous study, the linlog(linear log) RASTA(J-RASTA) approach based on PLP was proposed to deal with both the channel effect and the additive noise. The extraction of PLP required generally more steps and computation than the extraction of widely used MFCC. Thus, in this paper, we apply the linlog function to the MFCC for investigating the possibility of simple compensation method that removes both distortion. With the experimental results, the proposed method shows the similar tendency to the linlog RASTA-PLP_ When the J value is set to le-6, the best ERR(Error Reduction Rate) of 33% is obtained. For applying the linlog function to the feature extraction process, the J value plays a very important role in compensating the corruption. Thus, the study for the adaptive J or noise dependent J estimation is further required.

For speech recognition systems, rejection function as well as decoding function is necessary to improve the reliability. There have been many research efforts on out-of-vocabulary word rejection, however, little attention has been paid on non-target sentence rejection. Recently pronunciation approaches using speech recognition increase the need for non-target sentence rejection to provide more accurate and robust results. In this paper, we proposed filler model method and word/phoneme detection ratio method to implement non-target sentence rejection system. We made performance evaluation of filler model along to word-level, phoneme-level, and sentence-level filler models respectively. We also perform the similar experiment using word-level and phoneme-level word/phoneme detection ratio method. For the performance evaluation, the minimized average of FAR and FRR is used for comparing the effectiveness of each method along with the number of words of given sentences. From the experimental results, we got to know that word-level method outperforms the other methods, and word-level filler mode shows slightly better results than that of word detection ratio method.

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.

Among the conventional time-scale modification (TSM) methods, the synchronized overlap and add (SOLA) method is widely used due to its good performance relative to computational complexity But the SOLA method remains complex due to its synchronization procedure using the normalized cross-correlation function. In this paper, we introduce a computationally efficient SOLA method utilizing 3 level center clipping method, as well as zero-crossing and level-crossing information. The result of subjective preference test indicates that the proposed method can reduce the computational complexity by over 80% compared with the conventional SOLA method without serious degradation of synthesized speech quality.

This paper proposes a new structural head-related transfer function(HRTF) model to produce sounds in a virtual environment. The proposed HRTF model generates 3-D sounds by using a head model, a pinna model and the proposed distance model for azimuth, elevation, and distance that are three aspects for 3-D sounds, respectively. In particular, the proposed distance model consists of level normalization block distal region model, and proximal region model. To evaluate the performance of the proposed model, we setup an experimental procedure that each listener identifies a distance of 3-D sound sources that are generated by the proposed method with a predefined distance. It is shown from the tests that the proposed model provides an average distance error of $0.13{\sim}0.31$ meter when the sound source is generated as if it is 0.5 meter $\sim$ 2 meters apart from the listeners. This result is comparable to the average distance error of the human listening for the actual sound source.