• Proceedings of the IEEK Conference
• /
• 2003.07e
• /
• pp.2164-2167
• /
• 2003

In this paper we propose effective speech recognizer through recognition experiments for three feature parameters(PCA, ICA and MFCC) using SVM(Support Vector Machine) classifier In general, SVM is classification method which classify two class set by finding voluntary nonlinear boundary in vector space and possesses high classification performance under few training data number. In this paper we compare recognition result for each feature parameter and propose ICA feature as the most effective parameter

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.10
• /
• pp.37-43
• /
• 2021

Using the acoustic features of speech, important social and linguistic information about the speaker can be obtained, and one of the key features is the dialect. A speaker's use of a dialect is a major barrier to interaction with a computer. Dialects can be distinguished at various levels such as phonemes, syllables, words, phrases, and sentences, but it is difficult to distinguish dialects by identifying them one by one. Therefore, in this paper, we propose a lightweight Korean dialect classification model using only MFCC among the features of speech data. We study the optimal method to utilize MFCC features through Korean conversational voice data, and compare the classification performance of five Korean dialects in Gyeonggi/Seoul, Gangwon, Chungcheong, Jeolla, and Gyeongsang in eight machine learning and deep learning classification models. The performance of most classification models was improved by normalizing the MFCC, and the accuracy was improved by 1.07% and F1-score by 2.04% compared to the best performance of the classification model before normalizing the MFCC.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.24 no.1
• /
• pp.90-95
• /
• 2014

The researchs using brain-computer interface, the new interface system which connect human to macine, have been maded to implement the user-assistance devices for control of wheelchairs or input the characters. In recent researches, there are several trials to implement the speech recognitions system based on the brain wave and attempt to silent communication. In this paper, we studied how to extract features of vowel based on international phonetic alphabet (IPA), as a foundation step for implementing of speech recognition system based on electroencephalogram (EEG). We conducted the 2 step experiments with three healthy male subjects, and first step was speaking imagery with single vowel and second step was imagery with successive two vowels. We selected 32 channels, which include frontal lobe related to thinking and temporal lobe related to speech function, among acquired 64 channels. Eigen value of the signal was used for feature vector and support vector machine (SVM) was used for classification. As a result of first step, we should use over than 10th order of feature vector to analyze the EEG signal of speech and if we used 11th order feature vector, the highest average classification rate was 95.63 % in classification between /a/ and /o/, the lowest average classification rate was 86.85 % with /a/ and /u/. In the second step of the experiments, we studied the difference of speech imaginary signals between single and successive two vowels.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.64 no.9
• /
• pp.1363-1368
• /
• 2015

Recently, computer form were smaller than before because of computing technique's development and many wearable device are formed. So, computer's cognition of human emotion has importantly considered, thus researches on analyzing the state of emotion are increasing. Human voice includes many information of human emotion. This paper proposes a discriminative feature vector selection for emotion classification based on speech. For this, we extract some feature vectors like Pitch, MFCC, LPC, LPCC from voice signals are divided into four emotion parts on happy, normal, sad, angry and compare a separability of the extracted feature vectors using Bhattacharyya distance. So more effective feature vectors are recommended for emotion classification.

• Speech Sciences
• /
• v.9 no.3
• /
• pp.25-34
• /
• 2002

One of the most important factor in the performance of Korean text-to-speech system is the prediction of accentual and intonational phrase boundary. The previous method of prediction shows only the 75-85% which is not proper in the practical and commercial system. Therefore, more accurate prediction must be needed in the practical system. In this study, we propose the simple and more accurate method of the prediction of AP, IP.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1987.11a
• /
• pp.86-88
• /
• 1987

Accurate detection of the voicved segment in speech signals is important for robust pitch extraction. This paper describes an explicit detection algorithmfor detecting the voiced segment in speech signals. Thsi algoithm is based on the fluctuation properties of maximum magnitudes in each frame of speech signals. The performance of this detector is evaluated and compared to that obtained from manually classifying 150 recorded digit utterances.

• Speech Sciences
• /
• v.3
• /
• pp.148-155
• /
• 1998

As an application of dimensional analysis in the theory of chaos and fractals, we studied and estimated the information dimension for various phonemes. By constructing phase-space vectors from the time-series speech signals, we calculated the natural measure and the Shannon's information from the trajectories. The information dimension was finally obtained as the slope of the plot of the information versus space division order. The information dimension showed that it is so sensitive to the waveform and time delay. By averaging over frames for various phonemes, we found the information dimension ranges from 1.2 to 1.4.

• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.33-38
• /
• 2018

Although deep neural network-based acoustic models have greatly improved the performance of automatic speech recognition (ASR), reverberation still degrades the performance of distant speech recognition in indoor environments. In this paper, we adopt the DenseNet, which has shown great performance results in image classification tasks, to improve the performance of reverberant speech recognition. The DenseNet enables the deep convolutional neural network (CNN) to be effectively trained by concatenating feature maps in each convolutional layer. In addition, we extend the concept of multi-resolution CNN to multi-resolution DenseNet for robust speech recognition in reverberant environments. We evaluate the performance of reverberant speech recognition on the single-channel ASR task in reverberant voice enhancement and recognition benchmark (REVERB) challenge 2014. According to the experimental results, the DenseNet-based acoustic models show better performance than do the conventional CNN-based ones, and the multi-resolution DenseNet provides additional performance improvement.

• The KIPS Transactions:PartB
• /
• v.9B no.2
• /
• pp.231-236
• /
• 2002

The classification of the speech-sound block comes from the multi-resolution analysis property of the discrete wavelet transform, which is used to reduce the computational time for the pre-processing of speech recognition. The merging algorithm is proposed to extract vapid speech-sounds in terms of position and frequency range. It performs unvoiced/voiced classification and denoising. Since the merging algorithm can decide the processing parameters relating to voices only and is independent of system noises, it is useful for extracting valid speech-sounds. The merging algorithm has an adaptive feature for arbitrary system noises and an excellent denoising signal-to-noise ratio and a useful system tuning for the system implementation.

• ETRI Journal
• /
• v.38 no.2
• /
• pp.366-375
• /
• 2016

This paper describes an algorithm to suppress composite noise in a two-microphone speech enhancement system for robust hands-free speech communication. The proposed algorithm has four stages. The first stage estimates the power spectral density of the residual stationary noise, which is based on the detection of nonstationary signal-dominant time-frequency bins (TFBs) at the generalized sidelobe canceller output. Second, speech-dominant TFBs are identified among the previously detected nonstationary signal-dominant TFBs, and power spectral densities of speech and residual nonstationary noise are estimated. In the final stage, the bin-wise output signal-to-noise ratio is obtained with these power estimates and a Wiener post-filter is constructed to attenuate the residual noise. Compared to the conventional beamforming and post-filter algorithms, the proposed speech enhancement algorithm shows significant performance improvement in terms of perceptual evaluation of speech quality.