• The Journal of the Acoustical Society of Korea
• /
• v.27 no.1E
• /
• pp.1-10
• /
• 2008

A noise suppression algorithm is proposed for nonstationary noisy environments. The proposed algorithm is different from the conventional approaches such as the spectral subtraction algorithm and the minimum statistics noise estimation algorithm in that it classifies speech and noise signals in time-frequency bins. It calculates the ratio of the variance of the noisy power spectrum in time-frequency bins to its normalized time-frequency average. If the ratio is greater than an adaptive threshold, speech is considered to be present. Our adaptive algorithm tracks the threshold and controls the trade-off between residual noise and distortion. The estimated clean speech power spectrum is obtained by a modified gain function and the updated noisy power spectrum of the time-frequency bin. This new algorithm has the advantages of simplicity and light computational load for estimating the noise. This algorithm reduces the residual noise significantly, and is superior to the conventional methods.

• Proceedings of the KOSOMBE Conference
• /
• v.1990 no.05
• /
• pp.57-61
• /
• 1990

The purpose of this study is to help the hearing impaired's speech training with a visible speech processing system. In brief, this system converts the features of speech signals into graphics on monitor, and adjusts the features of hearing impaired to normal ones. There are form ant and pitch in the features used for this system. They are extracted using the digital signal processing such as linear prediotive method or AMDF(Average Magnitude Difference Function). In order to effectively train for the hearing impaired's abnormal speech, easilly visible feature has been being studied.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.5
• /
• pp.27-34
• /
• 2000

This paper deals with speech enhancement methods using the wavelet transform. A cycle-spinning scheme and undecimated wavelet transform are used for denoising of speech signals, and then their results are compared with that of the conventional wavelet transform. We apply soft-thresholding technique for removing additive background noise from noisy speech. The symlets 8-tap wavelet and pyramid algorithm are used for the wavelet transform. Performance assessments based on average SNR, cepstral distance and informal subjective listening test are carried out. Experimental results demonstrate that both cycle-spinning denoising(CSD) method and undecimated wavelet denoising(CWD) method outperform conventional wavelet denoising(UWD) method in objective performance measure as welt as subjective listening test. The two methods also show less "clicks" that usually appears in the neighborhood of signal discontinuities.

• Speech Sciences
• /
• v.12 no.4
• /
• pp.17-29
• /
• 2005

Time-frequency representation reveals some useful information about instantaneous frequency, instantaneous bandwidth and boundary of each AM-FM component of a speech signal. In many cases, the instantaneous frequency of each component is not constant. The variability of instantaneous frequency causes degradation of resolution in time-frequency representation. This paper presents a method of adaptively adjusting the transform kernel for preventing degradation of resolution due to time-varying instantaneous frequency. The transform kernel is the form of frequency modulated function. The modulation function in the transform kernel is determined by the estimate of instantaneous frequency which is approximated by first order polynomial at each time instance. Also, the window function is modulated by the estimated instantaneous. frequency for mitigation of fringing. effect. In the proposed method, not only the transform kernel but also the shape and the length of. the window function are adaptively adjusted by the instantaneous frequency of a speech signal.

• The Journal of the Acoustical Society of Korea
• /
• v.21 no.4
• /
• pp.156-156
• /
• 2002

We apply independent component analysis (ICA) for extracting an optimal basis to the problem of finding efficient features for representing speech signals of a given speaker The speech segments are assumed to be generated by a linear combination of the basis functions, thus the distribution of speech segments of a speaker is modeled by adapting the basis functions so that each source component is statistically independent. The learned basis functions are oriented and localized in both space and frequency, bearing a resemblance to Gabor wavelets. These features are speaker dependent characteristics and to assess their efficiency we performed speaker identification experiments and compared our results with the conventional Fourier-basis. Our results show that the proposed method is more efficient than the conventional Fourier-based features in that they can obtain a higher speaker identification rate.

• Journal of the Korean Institute of Telematics and Electronics
• /
• v.24 no.3
• /
• pp.381-386
• /
• 1987

By the speech production model, the first positive peak in a pitch interval of the voiced speech is mainly affected by the glottis and the first formant component, known as a typical energy source of the voiced speech. From these characteristics, the energy parameter can be replaced by the area of the area of the positve peak in a pitch interval, which parameter is generally used for classification of speech signals. In this method, the changed energy parameter is independent of window length applied for analysis, and the pitch can be extracted smultaneously. Furthermore, the energy can be extracted in the pitch period unit.

• Phonetics and Speech Sciences
• /
• v.10 no.1
• /
• pp.39-48
• /
• 2018

A typical statistical parametric speech synthesis (text-to-speech, TTS) system consists of separate modules, such as a text analysis module, an acoustic modeling module, and a speech synthesis module. This causes two problems: 1) expert knowledge of each module is required, and 2) errors generated in each module accumulate passing through each module. An end-to-end TTS system could avoid such problems by synthesizing voice signals directly from an input string. In this study, we implemented an end-to-end Korean TTS system using Google's Tacotron, which is an end-to-end TTS system based on a sequence-to-sequence model with attention mechanism. We used 4392 utterances spoken by a Korean female speaker, an amount that corresponds to 37% of the dataset Google used for training Tacotron. Our system obtained mean opinion score (MOS) 2.98 and degradation mean opinion score (DMOS) 3.25. We will discuss the factors which affected training of the system. Experiments demonstrate that the post-processing network needs to be designed considering output language and input characters and that according to the amount of training data, the maximum value of n for n-grams modeled by the encoder should be small enough.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.8 no.5
• /
• pp.665-670
• /
• 2013

In this paper, we propose a new method of speech detection and phoneme segmentation. We investigate the effect of applying preemphasis FIR filtering on the speech signal before the usual speech detection that utilizes the energy profile for discriminating signals from background noise. By this procedure, only the speech section of low energy and frequency becomes distinct in energy profile. It is verified experimentally that the silence/speech boundary becomes sharper by applying the filtering compared to the conventional method. By applications of this procedure, phoneme segmentation is also found to be much facilitated.

• Speech Sciences
• /
• v.7 no.2
• /
• pp.193-208
• /
• 2000

This paper describes a multi-step method of intonation modeling for corpus-based Korean speech synthesizer. We selected 1833 sentences considering various syntactic structures and built a corresponding speech corpus uttered by a female announcer. We detected the pitch using laryngograph signals and manually marked the prosodic boundaries on recorded speech, and carried out the tagging of part-of-speech and syntactic analysis on the text. The detected pitch was separated into 3 frequency bands of low, mid, high frequency components which correspond to the baseline, the word tone, and the syllable tone. We predicted them using the CART method and the Viterbi search algorithm with a word-tone-dictionary. In the collected spoken sentences, 1500 sentences were trained and 333 sentences were tested. In the layer of word tone modeling, we compared two methods. One is to predict the word tone corresponding to the mid-frequency components directly and the other is to predict it by multiplying the ratio of the word tone to the baseline by the baseline. The former method resulted in a mean error of 12.37 Hz and the latter in one of 12.41 Hz, similar to each other. In the layer of syllable tone modeling, it resulted in a mean error rate less than 8.3% comparing with the mean pitch, 193.56 Hz of the announcer, so its performance was relatively good.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.461-467
• /
• 2020

In noisy and multi-speaker environments, the performance of speech recognition is unavoidably lower than in a clean environment. To improve speech recognition, in this paper, the signal of the speaker of interest is extracted from the mixed speech signals with multiple speakers. The VoiceFilter model is used to effectively separate overlapped speech signals. In this work, clustering by Probabilistic Linear Discriminant Analysis (PLDA) similarity score was employed to detect the speech signal of the interested speaker, which is used as the reference speaker to VoiceFilter-based separation. Therefore, by utilizing the speaker feature extracted from the detected speech by the proposed clustering method, this paper propose a speaker diarization system using only the mixed speech without an explicit reference speaker signal. We use phone-dataset consisting of two speakers to evaluate the performance of the speaker diarization system. Source to Distortion Ratio (SDR) of the operator (Rx) speech and customer speech (Tx) are 5.22 dB and -5.22 dB respectively before separation, and the results of the proposed separation system show 11.26 dB and 8.53 dB respectively.