• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.6 no.2
• /
• pp.330-338
• /
• 2002

A speech recognition chip that can recognize a small vocabulary as a word-level has been designed. It is composed of EPD(Start and End-point detection) block, LPC block, DTW block and external memory interface block. It is made of 126,938 gates on 4x4mm2 area with a CMOS 0.35um TLM process. The speed of the chip varies from 5MHz to 60MHz because of its specific hardware designed for the purpose. It can compare 100,000 voices as a small vocabulary which has approximately 50∼60 frames at the clock of 5MHz and also up to 1,200,000 voices at the clock of 60MHz.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.359-366
• /
• 2022

In this paper, we present the Korean menu-ordering Sentence Text-to-Speech (TTS) system using conformer-based FastSpeech2. Conformer is the convolution-augmented transformer, which was originally proposed in Speech Recognition. Combining two different structures, the Conformer extracts better local and global features. It comprises two half Feed Forward module at the front and the end, sandwiching the Multi-Head Self-Attention module and Convolution module. We introduce the Conformer in Korean TTS, as we know it works well in Korean Speech Recognition. For comparison between transformer-based TTS model and Conformer-based one, we train FastSpeech2 and Conformer-based FastSpeech2. We collected a phoneme-balanced data set and used this for training our models. This corpus comprises not only general conversation, but also menu-ordering conversation consisting mainly of loanwords. This data set is the solution to the current Korean TTS model's degradation in loanwords. As a result of generating a synthesized sound using ParallelWave Gan, the Conformer-based FastSpeech2 achieved superior performance of MOS 4.04. We confirm that the model performance improved when the same structure was changed from transformer to Conformer in the Korean TTS.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.1
• /
• pp.86-91
• /
• 1999

This paper presents the text-prompt method to overcome the weakness of text-dependent and text-independent speaker recognition. Enhanced dynamic time warping for speaker recognition algorithm is applied. For the real-time processing, we use a simple algorithm for end-point detection without increasing computational complexity. The test shows that the weighted-cepstrum is most proper for speaker recognition among various speech parameters. As the experimental results of the proposed algorithm for three prompt words, the speaker identification error rate is 0.02%, and when the threshold is set properly, false rejection rate is 1.89%, false acceptance rate is 0.77% and verification total error rate is 0.97% for speaker verification.

• The Journal of the Acoustical Society of Korea
• /
• v.34 no.4
• /
• pp.321-327
• /
• 2015

In this paper, we propose an algorithm for achieving robust Visual Voice Activity Detection (VVAD) for enhanced speech recognition. In conventional VVAD algorithms, the motion of lip region is found by applying an optical flow or Chaos inspired measures for detecting visual speech frames. The optical flow-based VVAD is difficult to be adopted to driving scenarios due to its computational complexity. While invariant to illumination changes, Chaos theory based VVAD method is sensitive to motion translations caused by driver's head movements. The proposed Local Variance Histogram (LVH) is robust to the pixel intensity changes from both illumination change and translation change. Hence, for improved performance in environmental changes, we adopt the novel threshold estimation using total variance change. In the experimental results, the proposed VVAD algorithm achieves robustness in various driving situations.

• Journal of the Korea Convergence Society
• /
• v.8 no.12
• /
• pp.55-62
• /
• 2017

The Named Entity Recognition system is a system that recognizes words or phrases with object names such as personal name (PS), place name (LC), and group name (OG) in the document as corresponding object names. Traditional approaches to named entity recognition include statistical-based models that learn models based on hand-crafted features. Recently, it has been proposed to construct the qualities expressing the sentence using models such as deep-learning based Recurrent Neural Networks (RNN) and long-short term memory (LSTM) to solve the problem of sequence labeling. In this research, to improve the performance of the Korean named entity recognition system, we used a hand-crafted feature, part-of-speech tagging information, and pre-built lexicon information to augment features for representing sentence. Experimental results show that the proposed method improves the performance of Korean named entity recognition system. The results of this study are presented through github for future collaborative research with researchers studying Korean Natural Language Processing (NLP) and named entity recognition system.

• The Journal of the Acoustical Society of Korea
• /
• v.25 no.7
• /
• pp.312-318
• /
• 2006

The conventional lip-synch system has a two-step process, speech segmentation and recognition. However, the difficulty of speech segmentation procedure and the inaccuracy of training data set due to the segmentation lead to a significant Performance degradation in the system. To cope with that, the connected vowel recognition method using Head-Body-Tail (HBT) model is proposed. The HBT model which is appropriate for handling relatively small sized vocabulary tasks reflects co-articulation effect efficiently. Moreover the 7 vowels are merged into 3 classes having similar lip shape while the system is optimized by employing a class dependent SCHMM structure. Additionally in both end sides of each word which has large variations, 8 components Gaussian mixture model is directly used to improve the ability of representation. Though the proposed method reveals similar performance with respect to the CHMM based on the HBT structure. the number of parameters is reduced by 33.92%. This reduction makes it a computationally efficient method enabling real time operation.

• Journal of Digital Convergence
• /
• v.13 no.6
• /
• pp.151-156
• /
• 2015

In this paper, we have proposed personal multimodal biometric authentication system based on face detection, recognition and speaker verification for smart-phone environment. Proposed system detect the face with Modified Census Transform algorithm then find the eye position in the face by using gabor filter and k-means algorithm. Perform preprocessing on the detected face and eye position, then we recognize with Linear Discriminant Analysis algorithm. Afterward in speaker verification process, we extract the feature from the end point of the speech data and Mel Frequency Cepstral Coefficient. We verified the speaker through Dynamic Time Warping algorithm because the speech feature changes in real-time. The proposed multimodal biometric system is to fuse the face and speech feature (to optimize the internal operation by integer representation) for smart-phone based real-time face detection, recognition and speaker verification. As mentioned the multimodal biometric system could form the reliable system by estimating the reasonable performance.

• Journal of Korea Multimedia Society
• /
• v.4 no.4
• /
• pp.306-315
• /
• 2001

In noise reduction method from noisy speech for speech recognition in noisy environments, conventional spectral subtraction method has a disadvantage which distinction of noise and speech is difficult, and characteristic of noise can't be estimated accurately. Also, noise reduction method in the wavelet transform domain has a disadvantage which loss of signal is generated in the high frequency domain. In order to compensate theme disadvantage, this paper propose spectral subtraction method in continuous wavelet transform domain which speech and non- speech intervals is distinguished by standard deviation of wavelet coefficient, and signal is divided three scales at different scale. The proposed method extract accurately characteristic of noise in order to apply spectral subtraction method by end detection and band division. The proposed method shows better performance than noise reduction method using conventional spectral subtraction and wavelet transform from viewpoint signal to noise ratio and Itakura-Saito distance by experimental.

• Journal of the Korea Convergence Society
• /
• v.10 no.11
• /
• pp.33-39
• /
• 2019

Edutech, which combines education and information technology, is in the spotlight. Core technologies of the 4th Industrial Revolution have been actively used in education. Students use an AI-based learning platform to self-diagnose their needs. And get personalized training online with a cloud learning platform. Recently, a new educational medium called smart speaker that combines artificial intelligence technology and voice recognition technology has emerged and provides various educational services. The purpose of this study is to suggest a way to use smart speaker educationally to overcome the limitation of existing education. To this end, the concept and characteristics of smart speakers were analyzed, and the implications were derived by analyzing the contents provided by smart speakers. Also, the problem of using smart speaker was considered.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.109-117
• /
• 2021

Recently, the use of speech-based interfaces is increasing as a means for human-computer interaction (HCI). Accordingly, interest in post-processors for correcting errors in speech recognition results is also increasing. However, a lot of human-labor is required for data construction. in order to manufacture a sequence to sequence (S2S) based speech recognition post-processor. To this end, to alleviate the limitations of the existing construction methodology, a new data construction method called Back TranScription (BTS) was proposed. BTS refers to a technology that combines TTS and STT technology to create a pseudo parallel corpus. This methodology eliminates the role of a phonetic transcriptor and can automatically generate vast amounts of training data, saving the cost. This paper verified through experiments that data should be constructed in consideration of text style and domain rather than constructing data without any criteria by extending the existing BTS research.