• 한국음향학회지
• /
• 제27권8호
• /
• pp.418-426
• /
• 2008

본 논문에서는 스펙트로그램에서 마루와 골을 이용한 주파수 채널 분할 알고리즘을 제안한다. 주파수 채널 분할 문제는 동일한 음원으로부터 발생한 음성이 포함된 주파수 채널들을 하나의 그룹으로 묶는 것을 의미한다. 제안된 알고리즘은 입력 신호의 평탄화된 스펙트럼에 기반한 알고리즘이다. 평탄화된 스펙트럼에서 마루와 골은 각각 세그먼트의 중심과 경계를 판단하기 위해 사용된다. 각 세그먼트를 하나의 소리로 묶는 그룹핑 단계 이전에 제안된 알고리즘에 의한 세그멘테이션 결과가 유용함을 평가하기 위하여 이상적인 마스크에 의한 세그멘테이션 결과와 제안된 방법을 비교한다. 제안된 방법을 협대역 잡음, 광대역 잡음, 다른 음성신호와 혼합된 음성신호에 대하여 실험하였다.

• 한국전자통신학회논문지
• /
• 제17권3호
• /
• pp.409-416
• /
• 2022

본 논문은 시간지연 컨볼루션 혼합신호의 암묵잡음분리방법에 관한 것이다. 폐쇄된 공간에서 음향신호의 혼합모델은 다채널이기 때문에 convolutive 암묵신호분리방법을 적용하며 두 마이크 입력신호의 시간지연된 데이터 샘플들을 사용한다. 이 신호분리방법은 분리계수를 직접 계산하는 것이 아니라 역방향 모델을 이용하여 혼합계수를 산출하며, 계수의 갱신이 2차 통계적 성질에 기반한 반복적인 계산에 의해 이루어진다. 제안한 암묵신호분리의 성능을 검증하기 위해 많은 시뮬레이션을 수행하였다. 모의실험 결과, 이 방법을 사용한 잡음분리는 컨볼루션혼합에 상관없이 안전하게 동작하고, 일반적인 적응 FIR(Finite Impulse Response) 필터구조에 비해 PESQ(Perceptual Evaluation of Speech Quality)가 0.3점 개선되는 것으로 나타났다.

• 음성과학
• /
• 제10권2호
• /
• pp.7-25
• /
• 2003

Predicting the right prosodic elements is a key factor in improving the quality of synthesized speech. Prosodic elements include break, pitch, duration and loudness. Pitch, which is realized by Fundamental Frequency (F0), is the most important element relating to the quality of the synthesized speech. However, the previous method for predicting the F0 appears to reveal some problems. If voiced and unvoiced sounds are not correctly classified, it results in wrong prediction of pitch, wrong unit of triphone in synthesizing the voiced and unvoiced sounds, and the sound of click or vibration. This kind of feature is usual in the case of the transformation from the voiced sound to the unvoiced sound or from the unvoiced sound to the voiced sound. Such problem is not resolved by the method of grammar, and it much influences the synthesized sound. Therefore, to steadily acquire the correct value of pitch, in this paper we propose a new model for predicting and classifying the voiced and unvoiced sounds using the CART tool.

• 지능정보연구
• /
• 제29권3호
• /
• pp.267-286
• /
• 2023

인공지능 스피커로 대표되는 대화형 에이전트는 사람-컴퓨터 간 대화형이기 때문에 대화 상황에서 오류가 발생하는 경우가 잦다. 에이전트 사용자의 발화 기록에서 인식오류는 사용자의 발화를 제대로 인식하지 못하는 미인식오류 유형과 발화를 인식하여 서비스를 제공하였으나 사용자가 의도한 바와 다르게 인식된 오인식오류 유형으로 나뉜다. 이 중 오인식오류의 경우, 서비스가 제공된 것으로 기록되기 때문에 이에 대한 오류 탐지가 별도로 필요하다. 본 연구에서는 텍스트 마이닝 기법 중에서도 단어와 문서를 벡터로 바꿔주는 단어 임베딩과 문서 임베딩을 이용하여 단순 사용된 단어 기반의 유사도 산출이 아닌 단어의 분리 방식을 다양하게 적용함으로써 연속 발화 쌍의 유사도를 기반으로 새로운 오인식오류 및 신조어 탐지 방법을 탐구하였다. 연구 방법으로는 실제 사용자 발화 기록을 활용하여 오인식오류의 패턴을 모델 학습 및 생성 시 적용하여 탐지 모델을 구현하였다. 그 결과, 오인식오류의 가장 큰 원인인 등록되지 않은 신조어 사용을 탐지할 수 있는 패턴 방식으로 다양한 단어 분리 방식 중 초성 추출 방식이 가장 좋은 결과를 보임을 확인하였다. 본 연구는 크게 두 개의 함의를 가진다. 첫째, 인식오류로 기록되지 않아 탐지가 어려운 오인식오류에 대하여 다양한 방식 별 비교를 통해 최적의 방식을 찾았다. 둘째, 이를 실제 신조어 탐지 적용이 필요한 대화형 에이전트나 음성 인식 서비스에 적용한다면 음성 인식 단계에서부터 발생하는 오류의 패턴도 구체화할 수 있으며, 오류로 분류되지 않더라도 사용자가 원하는 결과에 맞는 서비스가 제공될 수 있음을 보였다.

• ETRI Journal
• /
• 제36권5호
• /
• pp.772-782
• /
• 2014

Two-microphone binary mask speech enhancement (2mBMSE) has been of particular interest in recent literature and has shown promising results. Current 2mBMSE systems rely on spatial cues of speech and noise sources. Although these cues are helpful for directional noise sources, they lose their efficiency in diffuse noise fields. We propose a new system that is effective in both directional and diffuse noise conditions. The system exploits two features. The first determines whether a given time-frequency (T-F) unit of the input spectrum is dominated by a diffuse or directional source. A diffuse signal is certainly a noise signal, but a directional signal could correspond to a noise or speech source. The second feature discriminates between T-F units dominated by speech or directional noise signals. Speech enhancement is performed using a binary mask, calculated based on the proposed features. In both directional and diffuse noise fields, the proposed system segregates speech T-F units with hit rates above 85%. It outperforms previous solutions in terms of signal-to-noise ratio and perceptual evaluation of speech quality improvement, especially in diffuse noise conditions.

• 음성과학
• /
• 제12권3호
• /
• pp.125-136
• /
• 2005

Automatic speech recognition is much more difficult in real world. Speech recognition according to SIR (Signal to Interface Ratio) is difficult in situations in which noise of surrounding environment and multi-speaker exists. Therefore, study on main speaker's voice extractions a very important field in speech signal processing in binaural sound. In this paper, we used directional filter and harmonic filter among other existing methods to extract the main speaker's information in binaural sound. The main speaker's voice was extracted using directional filter, and other remaining speaker's information was removed using harmonic filter through main speaker's pitch detection. As a result, voice of the main speaker was enhanced.

• 음성과학
• /
• 제12권4호
• /
• pp.53-70
• /
• 2005

A Vowel of speech signals are multicomponent signals composed of AM-FM components whose instantaneous frequency and instantaneous amplitude are time-varying. The changes of emotion states cause the variation of the instantaneous frequencies and the instantaneous amplitudes of AM-FM components. Therefore, it is important to estimate exactly the instantaneous frequencies and the instantaneous amplitudes of AM-FM components for the extraction of key information representing emotion states and changes in speech signals. In tills paper, firstly a method decomposing speech signals into AM - FM components is addressed. Secondly, the fundamental frequency of vowel sound is estimated by the simple method based on the spectrogram. The estimate of the fundamental frequency is used for decomposing speech signals into AM-FM components. Thirdly, an estimation method is suggested for separation of the instantaneous frequencies and the instantaneous amplitudes of the decomposed AM - FM components, based on Hilbert transform and the demodulation property of the extended Fourier transform. The estimates of the instantaneous frequencies and the instantaneous amplitudes can be used for modification of the spectral distribution and smooth connection of two words in the speech synthesis systems based on a corpus.

• 말소리와 음성과학
• /
• 제7권4호
• /
• pp.17-25
• /
• 2015

For blind source separation of convolutive mixtures, FDICA(Frequency Domain Independent Component Analysis) algorithms are generally used. Since FDICA algorithm such as Sawada FDICA, IVA(Independent Vector Analysis) works on the frequency bin basis with a natural gradient descent method, it takes much time to converge. In this paper, we propose a new method to improve convergence speed in FDICA algorithm. The proposed method reduces the number of iteration drastically in the process of natural gradient descent method by applying a weighted inner product constraint of unmixing matrix. Experimental results have shown that the proposed method achieved large improvement of convergence speed without degrading the separation performance of the baseline algorithms.

• The Journal of the Acoustical Society of Korea
• /
• 제27권1E호
• /
• pp.30-34
• /
• 2008

An improvement to the existing blind signal separation (BSS) method has been made in this paper. The proposed method models the inherent signal dependency observed in acoustic object to separate the real-world convolutive sound mixtures. The frequency domain approach requires solving the well known permutation problem, and the problem had been successfully solved by a vector representation of the sources whose multidimensional joint densities have a certain amount of dependency expressed by non-spherical distributions. Especially for speech signals, we observe strong dependencies across neighboring frequency bins and the decrease of those dependencies as the bins become far apart. The non-spherical joint density model proposed in this paper reflects this property of real-world speech signals. Experimental results show the improved performances over the spherical joint density representations.

• 디지털산업정보학회논문지
• /
• 제12권1호
• /
• pp.43-50
• /
• 2016

This paper is on an acoustic echo canceller solving the double-talk problem by using the blind signal separation technology. The acoustic echo canceller may be deteriorated or diverged during the double-talk period. So we use the blind signal separation to detect the double talking by separating the near-end speech signal from the mixed microphone signal. The blind signal separation extracts the near-end signal from dual microphones by the iterative computations using the 2nd order statistical character in the closed reverberation environment. By this method, the acoustic echo canceller operates irrespective of the double-talking. We verified performances of the proposed acoustic echo canceller in the computer simulations. The results show that the acoustic echo canceller with this algorithm detects the double-talk periods well, and then operates stably without diverging of the coefficients after ending the double-talking. The merits are in the simplicity and stability.