MobileNet을 이용한 사람 음성 구간의 오디오 축약 방법

이재준,한현택,최연웅,이해연 한국정보기술학회 2022 한국정보기술학회논문지 Vol.20 No.3

Applications for collecting and utilizing audio information from various smart devices are being developed. Human voice is important data among vast amounts of audio, and it is useful to deduct only voice activity segments from audio. This paper proposes a method to contract only voice activity segments excluding non-speech segments using MobileNet. Input audio is divided into 3 second segments and MFCC features are extracted and used for voice detection. A CNN model widely used in the past has a problem of increasing the amount of computation due to its deep structure. Therefore, MobileNet focused on optimizing the amount of computation is used. Experiments were performed using domestic and foreign datasets and audio collected by ourselves. As a result, we achieved the voice detection accuracy of 93.92% for each segment and the reduction accuracy of 88.05% for the entire audio. 다양한 스마트 기기에서 오디오 정보들을 수집하고 활용하는 응용들이 개발되고 있다. 방대한 오디오 중에서 사람 음성은 중요한 정보로 오디오에서 음성 구간만 축약하는 것은 유용하다. 본 논문에서는 MobileNet을 사용하여 오디오에서 비음성 구간들을 제외한 음성 구간만을 축약시키는 방법을 제안한다. 입력 오디오를 3초 단위 세그먼트로 구분하고, MFCC 특징을 추출하여 사람 음성 판별에 활용하였다. 특히, 기존에 많이 사용되는 CNN 모델은 구조가 깊어져서 연산량이 증가하는 문제가 있어서, 연산량 최적화에 중점을 둔 MobileNet을 활용하였다. 국내외 여러 데이터셋과 자체적으로 수집한 오디오를 사용하여 실험을 수행하였고, 그 결과 세그먼트 단위로 93.92% 음성 검출 정확도와 전체 오디오에 대해 88.05%의 축약 정확도를 달성하였다.

Robust Voice Activity Detection Using the Spectral Peaks of Vowel Sounds

유인철,육동석 한국전자통신연구원 2009 ETRI Journal Vol.31 No.4

This letter proposes the use of vowel sound detection for voice activity detection. Vowels have distinctive spectral peaks. These are likely to remain higher than their surroundings even after severe corruption. Therefore, by developing a method of detecting the spectral peaks of vowel sounds in corrupted signals, voice activity can be detected as well even in low signal-to-noise ratio (SNR) conditions. Experimental results indicate that the proposed algorithm performs reliably under various noise and low SNR conditions. This method is suitable for mobile environments where the characteristics of noise may not be known in advance.

Dempster-Shafer theory for enhanced statistical model-based voice activity detection

Park, Tae-Jun,Chang, Joon-Hyuk Elsevier 2018 Computer speech & language Vol.47 No.-

<P><B>Abstract</B></P> <P>In this paper, we propose to combine the posterior probabilities of voice activity derived from different statistical model-based algorithms for enhanced voice activity detection. For this, the Dempster-Shafer (DS) theory of evidence is employed to represent and combine the different probabilities estimated by three different statistical model-based VAD algorithms including the Sohn’s likelihood ratio test (LRT)-based method, smoothed LRT-based method, and multiple observation LRT-based method. By considering a generalization of the Bayesian framework and permitting the characterization of uncertainty and ignorance through the DS theory, the probability of an ignorant state is eliminated through the orthogonal sum of several speech presence probabilities, which results in the performance improvement when detecting voice activity. According to objective test results, it is discovered the proposed DS theory-based VAD method offers significant improvements over the conventional approaches.</P> <P><B>Highlights</B></P> <P> <UL> <LI> We develop the voice activity detection based on DS theory. </LI> <LI> Three statistical model-based VADs are used as the baseline systems. </LI> <LI> Probabilities from the three VADs are combined to DS theory. </LI> <LI> Proposed system works well over the existing methods. </LI> </UL> </P>

Applying the Bi-level HMM for Robust Voice-activity Detection

황용원,정문호,오상록,김일환 대한전기학회 2017 Journal of Electrical Engineering & Technology Vol.12 No.1

This paper presents a voice-activity detection (VAD) method for sound sequences with various SNRs. For real-time VAD applications, it is inadequate to employ a post-processing for the removal of burst clippings from the VAD output decision. To tackle this problem, building on the bilevel hidden Markov model, for which a state layer is inserted into a typical hidden Markov model (HMM), we formulated a robust method for VAD not requiring any additional post-processing. In the method, a forward-inference-ratio test was devised to detect the speech endpoints and Mel-frequency cepstral coefficients (MFCC) were used as the features. Our experiment results show that, regarding different SNRs, the performance of the proposed approach is more outstanding than those of the conventional methods.

Applying the Bi-level HMM for Robust Voice-activity Detection

Yongwon Hwang,Mun-Ho Jeong,Sang-Rok Oh,Il-Hwan Kim 대한전기학회 2017 Journal of Electrical Engineering & Technology Vol.12 No.1

This paper presents a voice-activity detection (VAD) method for sound sequences with various SNRs. For real-time VAD applications, it is inadequate to employ a post-processing for the removal of burst clippings from the VAD output decision. To tackle this problem, building on the bilevel hidden Markov model, for which a state layer is inserted into a typical hidden Markov model (HMM), we formulated a robust method for VAD not requiring any additional post-processing. In the method, a forward-inference-ratio test was devised to detect the speech endpoints and Mel-frequency cepstral coefficients (MFCC) were used as the features. Our experiment results show that, regarding different SNRs, the performance of the proposed approach is more outstanding than those of the conventional methods.

Applying the Bi-level HMM for Robust Voice-activity Detection

Hwang, Yongwon,Jeong, Mun-Ho,Oh, Sang-Rok,Kim, Il-Hwan The Korean Institute of Electrical Engineers 2017 Journal of Electrical Engineering & Technology Vol.12 No.1

This paper presents a voice-activity detection (VAD) method for sound sequences with various SNRs. For real-time VAD applications, it is inadequate to employ a post-processing for the removal of burst clippings from the VAD output decision. To tackle this problem, building on the bi-level hidden Markov model, for which a state layer is inserted into a typical hidden Markov model (HMM), we formulated a robust method for VAD not requiring any additional post-processing. In the method, a forward-inference-ratio test was devised to detect the speech endpoints and Mel-frequency cepstral coefficients (MFCC) were used as the features. Our experiment results show that, regarding different SNRs, the performance of the proposed approach is more outstanding than those of the conventional methods.

상태변수 기반의 실시간 음성검출 알고리즘의 최적화

김수환(Kim Suhwan),이영재(Lee Youngjae),김영일(Kim Young-Il),정상배(Jeong Sangbae) 한국음성학회 2010 말소리와 음성과학 Vol.2 No.4

In this paper, a speech endpoint detection algorithm is proposed. The proposed algorithm is a kind of state transition-based ones for speech detection. To reject short-duration acoustic pulses which can be considered noises, it utilizes duration information of all detected pulses. For the optimization of parameters related with pulse lengths and energy threshold to detect speech intervals, an exhaustive search scheme is adopted while speech recognition rates are used as its performance index. Experimental results show that the proposed algorithm outperforms the baseline state-based endpoint detection algorithm. At 5 ㏈ input SNR for the beamforming input, the word recognition accuracies of its outputs were 78.5% for human voice noises and 81.1% for music noises.

A Weighted Feature Voting Approach for Robust and Real-Time Voice Activity Detection

Mohammad Hossein Moattar,Mohammad Mehdi Homayounpour 한국전자통신연구원 2011 ETRI Journal Vol.33 No.1

This paper concerns a robust real-time voice activity detection (VAD) approach which is easy to understand and implement. The proposed approach employs several short-term speech/nonspeech discriminating features in a voting paradigm to achieve a reliable performance in different environments. This paper mainly focuses on the performance improvement of a recently proposed approach which uses spectral peak valley difference (SPVD) as a feature for silence detection. The main issue of this paper is to apply a set of features with SPVD to improve the VAD robustness. The proposed approach uses a weighted voting scheme in order to take the discriminative power of the employed feature set into account. The experiments show that the proposed approach is more robust than the baseline approach from different points of view, including channel distortion and threshold selection. The proposed approach is also compared with some other VAD techniques for better confirmation of its achievements. Using the proposed weighted voting approach, the average VAD performance is increased to 89.29% for 5 different noise types and 8 SNR levels. The resulting performance is 13.79% higher than the approach based only on SPVD and even 2.25% higher than the not-weighted voting scheme.

Efficient Implementation of Voiced/Unvoiced Sounds Classification Based on GMM for SMV Codec

SONG, Ji-Hyun,CHANG, Joon-Hyuk The Institute of Electronics, Information and Comm 2009 IEICE transactions on fundamentals of electronics, Vol.92 No.8

<P>In this letter, we propose an efficient method to improve the performance of voiced/unvoiced (V/UV) sounds decision for the selectable mode vocoder (SMV) of 3GPP2 using the Gaussian mixture model (GMM). We first present an effective analysis of the features and the classification method adopted in the SMV. And feature vectors which are applied to the GMM are then selected from relevant parameters of the SMV for the efficient V/UV classification. The performance of the proposed algorithm are evaluated under various conditions and yield better results compared to the conventional method of the SMV.</P>

Robust Entropy Based Voice Activity Detection Using Parameter Reconstruction in Noisy Environment

Han, Hag-Yong,Lee, Kwang-Seok,Koh, Si-Young,Hur, Kang-In The Korea Institute of Information and Commucation 2003 Journal of information and communication convergen Vol.1 No.4

Voice activity detection is a important problem in the speech recognition and speech communication. This paper introduces new feature parameter which are reconstructed by spectral entropy of information theory for robust voice activity detection in the noise environment, then analyzes and compares it with energy method of voice activity detection and performance. In experiments, we confirmed that spectral entropy and its reconstructed parameter are superior than the energy method for robust voice activity detection in the various noise environment.