• 한국음향학회지
• /
• 제40권2호
• /
• pp.176-182
• /
• 2021

본 논문은 다양한 손실 함수에 따른 Deep Nerual Network(DNN) 기반 음성 향상 모델의 성능을 비교 평가한다. 베이스라인 모델로는 음성의 위상 정보를 고려할 수 있는 복소 네트워크를 사용하였다. 손실 함수는 두 가지 유형의 기본 손실 함수, Mean Squared Error(MSE)와 Scale-Invariant Source-to-Noise Ratio(SI-SNR)를 사용하였으며 두 가지 유형의 지각 기반 손실 함수 Perceptual Metric for Speech Quality Evaluation(PMSQE)과 Log Mel Spectra(LMS)를 사용한다. 성능은 각 손실 함수의 다양한 조합을 사용하여 얻은 출력을 객관적인 평가와 청취 테스트를 통해 측정하였다. 실험 결과, 지각기반 손실 함수를 MSE 또는 SI-SNR과 결합하였을 때 전반적으로 성능이 향상되며, 지각기반 손실함수를 사용하면 객관적 지표에서 약세를 보이는 경우라도 청취 테스트에서 우수한 성능을 보임을 확인하였다.

• 대한음성학회지:말소리
• /
• 제50호
• /
• pp.41-50
• /
• 2004

Recently, there has been considerable progress in the natural language processing and digital signal processing components and this progress has led to the improved synthetic speech qualify of many commercial TTS systems. But there still remain many obstacles to overcome for the practical application of TTS. To resolve the problems, the cooperative research among the related areas is highly required and a common Korean TTS platform is essential to promote these activities. This platform offers a general framework for building Korean speech synthesis systems and a full C/C++ source for modules supports to implement and test his own algorithm. In this paper we described the aspect of a Korean TTS platform to be developed and a developing plan.

• ETRI Journal
• /
• 제32권6호
• /
• pp.921-931
• /
• 2010

It has been shown that the principal subspace-based multichannel Wiener filter (MWF) provides better performance than the conventional MWF for suppressing interference in the case of a single target source. It can efficiently estimate the target speech component in the principal subspace which estimates the acoustic transfer function up to a scaling factor. However, as the input signal-to-interference ratio (SIR) becomes lower, larger errors are incurred in the estimation of the acoustic transfer function by the principal subspace method, degrading the performance in interference suppression. In order to alleviate this problem, a principal subspace modification method was proposed in previous work. The principal subspace modification reduces the estimation error of the acoustic transfer function vector at low SIRs. In this work, a frequency-band dependent interpolation technique is further employed for the principal subspace modification. The speech recognition test is also conducted using the Sphinx-4 system and demonstrates the practical usefulness of the proposed method as a front processing for the speech recognizer in a distant-talking and interferer-present environment.

• The Journal of the Acoustical Society of Korea
• /
• 제21권4E호
• /
• pp.156-163
• /
• 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.

• 한국소음진동공학회:학술대회논문집
• /
• 한국소음진동공학회 2007년도 춘계학술대회논문집
• /
• pp.649-654
• /
• 2007

Eavesdropping prevention performance is evaluated on a commercial window shaker, which is used to prevent a glass window from eavesdropping. Speech transmission index (STI) is introduced in order to estimate quantitatively the speech intelligibility of the sound detected on the glass window. Objective test by IEC standard using modulation transfer function (MTF) is performed to determine STI. Using Maximum Length Sequency (MLS) signal as a sound source, MTF is measured by accelerometers and laser doppler vibrometer. STI under different level of disturbing wave are compared to confirm the disturbing effect on the speech intelligibility.

• 음성과학
• /
• 제10권2호
• /
• pp.237-248
• /
• 2003

This paper compares the syllable structure of French and Korean analyzing the speech data of these two languages recorded during the actual speech. Reference to the syllable structure of French is made from F. Wioland's research data. As for the Korean data, the primary data are drawn from the 30-minute radio interview in which two male TV anchors in their early 60s talk to each other. The secondary source of the data is collected by having the primary data replicated by the two male announcers in their early 20's broadcasting in the university ra야o station of KAIST. With reference to the data collected in French and Korean, this paper provides the statistical frequency of each type of syllable structure in each language through the acoustic analysis of the spectrograms and renders a phonetic account of the characteristics of each syllable type in the two languages. Also discussed in this paper is the distributional condition in which each syllable structure is laid out in the speech context.

• 음성과학
• /
• 제13권1호
• /
• pp.117-128
• /
• 2006

The two-channel Generalized Sidelobe Canceller (GSC) scheme suffers from the presence of leakage signal in the reference channel. The leakage signal is caused by the dissimilar impulse responses between microphones, and different paths from speech source to microphones. Such leakage is detrimental to speech enhancement of the GSC since the desired reference signal becomes corrupted. In order to suppress the signal leakage, two matrix injection methods are proposed. In the first method, a simple gain compensation matrix is used. In the second, a projection matrix for reducing the error between the actual and the ideal primary and reference signals, is used. This paper describes the performance degradation resulting from leakage, and proposes effective methods to resolve the problem. Representative experiments were conducted to demonstrate the effectiveness of the proposed methods on recorded speech and noise in an actual automobile environment.

• 한국정보처리학회논문지
• /
• 제7권4호
• /
• pp.1299-1304
• /
• 2000

In a speed coding system using excitation source of voiced and unvoiced, it would be involved a distortion of speech quality in case coexist with a voiced and an unvoiced consonants in a frame. So, I propose TSIUVC(Transition Segment Including UnVoiced Consonant) searching, extraction ad approximation-synthesis method in order to uncoexistent with a voiced and unvoiced consonants in a frame. This method based on a zerocrossing rate and pitch detector using FIR-STREAK Digital Filter. As a result, the extraction rates of TSIUVC are 84.8% (plosive), 94.9%(fricative), 92.3%(affricative) in female voice, and 88%(plosive), 94.9%(fricative), 92.3%(affricative) in male voice respectively, Also, I obain a high quality approximation-synthesis waveforms within TSIUVC by using frequency information of 0.547kHz below and 2.813kHz above. This method has the capability of being applied to speech coding of low bit rate, speech analysis and speech synthesis.

• 음성과학
• /
• 제10권4호
• /
• pp.137-147
• /
• 2003

Speech synthesis can classify by synthesis way, that is waveform coding, source coding and mixture coding. Specially, waveform coding is suitable for high quality synthesis. However, it is not desirable by synthesis techniques of syllable or phoneme unit because it do not separate and handles excitation and formant part. Therefore, there is a need for pitch alteration method applied in synthesis by the rule in waveform coding. This study propose about pitch alteration method that use spectrum scaling after do to flatten spectra by subband linear approximation to minimize spectrum distortion. This paper show evaluation whether show excellency of some measure compared with LPC, Cepstrum, lifter function and method that propose. estimation method seeks distribution of each flattened signal and measured degree of flattened spectra Signal flattened is normalized, So that highest point amounts to zero, and distribution of signal ,whose average is zero, is calculated. this show result that measure the spectrum distortion rate to estimate performance of method that propose. The average spectrum distortion rate was kept below the average 2.12%, so the method that propose is superiors than existent method.

• 한국음향학회지
• /
• 제21권4호
• /
• 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.