Spectral Subtraction Using Spectral Harmonics for Robust Speech Recognition in Car Environments

Beh, Jounghoon,Ko, Hanseok The Acoustical Society of Korea 2003 韓國音響學會誌 Vol.22 No.e2

This paper addresses a novel noise-compensation scheme to solve the mismatch problem between training and testing condition for the automatic speech recognition (ASR) system, specifically in car environment. The conventional spectral subtraction schemes rely on the signal-to-noise ratio (SNR) such that attenuation is imposed on that part of the spectrum that appears to have low SNR, and accentuation is made on that part of high SNR. However, these schemes are based on the postulation that the power spectrum of noise is in general at the lower level in magnitude than that of speech. Therefore, while such postulation is adequate for high SNR environment, it is grossly inadequate for low SNR scenarios such as that of car environment. This paper proposes an efficient spectral subtraction scheme focused specifically to low SNR noisy environment by extracting harmonics distinctively in speech spectrum. Representative experiments confirm the superior performance of the proposed method over conventional methods. The experiments are conducted using car noise-corrupted utterances of Aurora2 corpus.

Dual channel based speech enhancement using novelty filter for robust speech recognition in automobile environment

Beh, Jounghoon,Baran, R.H.,Ko, Hanseok IEEE 2006 IEEE transactions on consumer electronics Vol.52 No.2

In this paper, a simple and effective dual channel speech enhancement algorithm is proposed. The proposed algorithm incorporates a novelty filter scheme, performing associative memory and aiming at improving the performance of the dual channel generalized sidelobe canceller (GSC.) The novelty filter is demonstrated as effective for enhancing the reference noise used in the dual channel GSC, such that the assumptions required for adaptive noise cancellation are satisfied. In this paper, the concept of a novelty filter scheme is described, in addition to the conditions required to guarantee, the performance of the dual channel GSC and the method of applying the novelty filter scheme to speech enhancement objectives. In representative experiments, superiority of the proposed algorithm is demonstrated using three objective measures; segmental signal to noise ratio (SNR), log spectral distance (LSD) and word accuracy, which are evaluated via speech data recorded in real automobile environments.

An Energy-Efficient MAC Protocol for Wireless Wearable Computer Systems

Beh, Jounghoon,Hur, Kyeong,Kim, Wooil,Joo, Yang-Ick The Korea Institute of Information and Commucation 2013 Journal of information and communication convergen Vol.11 No.1

Wearable computer systems use the wireless universal serial bus (WUSB), which refers to USB technology that is merged with WiMedia physical layer and medium access control layer (PHY/MAC) technical specifications. WUSB can be applied to wireless personal area network (WPAN) applications as well as wired USB applications such as PAN. WUSB specifications have defined high-speed connections between a WUSB host and WUSB devices for compatibility with USB 2.0 specifications. In this paper, we focus on an integrated system with a WUSB over an IEEE 802.15.6 wireless body area network (WBAN) for wireless wearable computer systems. Due to the portable and wearable nature of wearable computer systems, the WUSB over IEEE 802.15.6 hierarchical medium access control (MAC) protocol has to support power saving operations and integrate WUSB transactions with WBAN traffic efficiently. In this paper, we propose a low-power hibernation technique (LHT) for WUSB over IEEE 802.15.6 hierarchical MAC to improve its energy efficiency. Simulation results show that the LHT also integrates WUSB transactions and WBAN traffic efficiently while it achieves high energy efficiency.

Parameter Estimation of Source Image Using Least-Squares for Dual Channel Underdetermined Convolutive Blind Speech Separation

Jounghoon Beh(배정훈) 한국산학기술학회 2021 한국산학기술학회논문지 Vol.22 No.10

칵테일 파티 문제를 해결하기 위한 일환으로서 블라인드 음원 분리 기술은 군사용 인공지능 활용 분야에서 오랫동안 연구되어왔다. 칵테일 파티 문제는 언어이해를 목적으로 하는 지능 시스템이 다양한 음원이 존재하는 상황에서 타겟사용자의 목소리를 분간해 내지 못하여 사용자 명령을 이해하지 못하는 현상을 말한다. 본 논문에서는 이중 채널 마이크로폰 어레이 상황에서 언더디터민드 (underdetermined) 하고 합성곱 형식으로 섞여진 신호의 혼합계수 (mixing parameter)를 추정하는 새로운 방법을 제안한다. 최소제곱법 원리를 이용하여 혼합계수들을 추정하는 방법이 소개되며, 분리된 신호는 마이크로폰에서의 신호 형상 (signal image) 레벨로 구해진다. 이 결과, 일반적으로 블라인드 음원분리 기술에서 맞닥뜨리는 스케일링 모호성 (scaling ambiguity)이 없어지게 된다. 제안한 방법은 최소제곱법을 기반을 두며, 일반적으로 블라인드 음원 분리 기술에서 맞닥뜨리는 스케일링 모호성 (scaling ambiguity)를 없앤다. 제안한 방법의 성능은 학계에 공개된 SiSEC 2008 데이터셑을 이용해 측정되었으며, 다양한 종류의 신호 대 잡음비를 측정하고, 그 측정값을 같은 분야의 최신 기술들과 비교함으로써 그 타당성을 입증하였다. Blind source separation for the cocktail-party problem has long been a challenge in Artificial Intelligence (AI) for military applications. The cocktail-party problem states that an intelligent machine cannot locate and listen to the target speaker"s voice when it is surrounded by other sound sources such as in a cocktail party. This paper proposes a novel method to estimate mixing parameters of underdetermined and convolutive mixture of speech sources given a dual-channel microphone array. In the proposed method, the optimal estimate of mixing parameters and the source image is obtained using the least-squares principle. As a result, the scaling ambiguity, which is common in conventional blind source separation methods, is alleviated. Performance evaluation of the proposed method is conducted with a public dataset, SiSEC 2008, and experimental results show the present method"s validity in terms of various SNR measures compared to other state-of-the-art techniques in the field.

Harmonics-based Spectral Subtraction and Feature Vector Normalization for Robust Speech Recognition

Beh, Jounghoon,Lee, Heungkyu,Kwon, Ohil,Ko, Hanseok 한국음성과학회 2004 음성과학 Vol.11 No.1

In this paper, we propose a two-step noise compensation algorithm in feature extraction for achieving robust speech recognition. The proposed method frees us from requiring a priori information on noisy environments and is simple to implement. First, in frequency domain, the Harmonics-based Spectral Subtraction (HSS) is applied so that it reduces the additive background noise and makes the shape of harmonics in speech spectrum more pronounced. We then apply a judiciously weighted variance Feature Vector Normalization (FVN) to compensate for both the channel distortion and additive noise. The weighted variance FVN compensates for the variance mismatch in both the speech and the non-speech regions respectively. Representative performance evaluation using Aurora 2 database shows that the proposed method yields 27.18% relative improvement in accuracy under a multi-noise training task and 57.94% relative improvement under a clean training task.

MAP Adaptation with Mixtures of von Mises Distributions and Its Application to Underdetermined Convolutive Blind Source Separation

Jounghoon Beh(배정훈) 한국산학기술학회 2021 한국산학기술학회논문지 Vol.22 No.10

본 논문에서는 시간-주파수 영역에서 음원분리를 위해 도달음원의 위상차 분포 추정을 위한 최대 사후 확률 기반의 파라미터 적응기법을 제안한다. 도달 음원의 위상차 데이터는 혼합 폰 미제스 분포로 가정하여 음원의 각도 위치에 따라 분류되었으며, 이 혼합 폰 미제스 분포는 위상차와 같은 2π 순환 데이터를 모델링할 때 효과적이다. 제안한 방법은 혼합 음원의 스펙트로그램 상에서 시간 및 주파수 구간을 각각의 음원으로 클러스터링하는 용도에 사용되었으며, 이는 마스크 기반 블라인드 음원 분리기술에서 가장 핵심적인 부분이다. 혼합 폰 미제스 분포로 모델링 된 도달음원위상차 분포의 파라미터들을 안정적으로 추정하기 위하여, 먼저 각 도착 신호의 각도 분포를 구성하였으며, 이 각도 분포를 바탕으로 각 주파수 밴드의 위상차 분포를 나타내는 파라미터들을 각각의 주파수 대역 별로 추정하였다. 이를 바탕으로 스펙트로그램상에서 음원별로 시간-주파수 영역을 구분하고 이 영역들은 시간-주파수 마스크를 생성하는데 쓰여졌다. 제안한 방법은 SiSEC 2008 데이터셑을 이용하여 다양한 신호 대 잡음비 값으로 평가되었으며, 그 타당성이 입증되었다. In this paper, we propose a Maximum a Posteriori (MAP) adaptation method for the parameter of Phase Difference of Arrival (PDOA) distribution to classify sound sources in the time-frequency domain. The PDOA is clustered via Mixtures of von Mises distribution (MovM) which is efficacious in modeling a 2 circular domain. The proposed method is employed to the task of clustering a spectral bin in terms of a sound source, which is a crucial part of the mask-based blind source separation. To robustly estimate the model parameter of MovM of the PDOA, we first build an incidence angle distribution and then adapt the parameters to each frequency band. The clustered spectral bin is used to build a time-frequency (TF) mask to separate the mixed audio signal based on the signal sparseness property.

Motion Primitives for Designing Flexible Gesture Set in Human–Robot Interface

Suwon Shon,Jounghoon Beh,Cheoljong Yang,David K. Han,Hanseok Ko 제어로봇시스템학회 2011 제어로봇시스템학회 국제학술대회 논문집 Vol.2011 No.10

This paper proposes motion primitives for designing a gesture set in a gesture recognition system as Human-Robot Interface (HRI). Based on statistical analyses of angular tendency of hand movements in sign languages and hand motions in practical gestures, we construct four motion primitives as building blocks for basic hand motions. By combining these motion primitives, we design a discernable "fundamental hand motion set" toward improving machine based hand signal recognition. Novelty of combining the proposed motion primitives is demonstrated by a "fundamental hand motion set" recognizer based on Hidden Markov Model (HMM). The recognition system shows 99.40% recognition rate on the proposed language set. For connected recognition of the "fundamental hand motion set", the recognition system shows 97.95% recognition rate. The results validate that using the proposed motion primitives ensures flexibility and discernability of a gesture set. It is thus promising candidate for standardization when designing gesture sets for human-robot interface.

수신호 인식기를 이용한 로봇 사용자 제어 시스템

손수원(Suwon Shon),배정훈(Jounghoon Beh),양철종(Cheoljong Yang),왕한(Han Wang),고한석(Hanseok Ko) 제어로봇시스템학회 2011 제어·로봇·시스템학회 논문지 Vol.17 No.4

This paper proposes a robot control human interface using Markov model (HMM) based hand signal recognizer. The command receiving humanoid robot sends webcam images to a client computer. The client computer then extracts the intended commanding human’ hand motion descriptors. Upon the feature acquisition, the hand signal recognizer carries out the recognition procedure. The recognition result is then sent back to the robot for responsive actions. The system performance is evaluated by measuring the recognition of ‘8 hand signal set’which is created randomly using fundamental hand motion set. For isolated motion recognition, ‘8 hand signal set’shows 97.07% recognition rate while the ‘aseline hand signal set’shows 92.4%. This result validates the proposed hand signal recognizer is indeed highly discernable. For the ‘8 hand signal set’connected motions, it shows 97.37% recognition rate. The relevant experiments demonstrate that the proposed system is promising for real world human-robot interface application.

영상기반의 안정적 수신호 인식기를 위한 손동작 패턴 설계 방법

손수원(Suwon Shon),배정훈(Jounghoon Beh),양철종(Cheoljong Yang),왕한(Han Wang),고한석(Hanseok Ko) 大韓電子工學會 2011 電子工學會論文誌-SP (Signal processing) Vol.48 No.4

본 논문에서는 수신호 인식기에 쓰이기 위한 분별성 있는 손동작을 만드는 방법을 제안한다. 기존의 수화DB에서 손의 움직임을 분석하여 기본 동작이 되는 4가지의 모션 프리미티브를 선정하였으며, 선정된 모션 프리미티브를 조합하여 구별성 있는 ‘기본 손동작 집합‘을 제작하였다. 제안하는 ‘기본 손동작 집합‘ 의 구별성을 증명하기 위하여 ’기본 손동작 집합‘ 인식기를 만들고 인식결과를 확인하였다. 사용된 인식기는 hidden Markov model (HMM) 을 기반으로 제작되었다. 기본 손동작 인식 task에 대한 성능평가 결과 99.01%로써 각 모델 간에 높은 구별성을 보이는 것을 확인할 수 있었다. This paper proposes a language set of hand motions for enhancing the performance of vision-based hand signal recognizer. Based on the statistical analysis of the angular tendency of hand movements in sign language and the hand motions in practical use, we construct four motion primitives as building blocks for basic hand motions. By combining these motion primitives, we design a discernable ‘fundamental hand motion set’ toward increasing the hand signal recognition. To demonstrate the validity of proposed designing method, we develop a ‘fundamental hand motion set’ recognizer based on hidden Markov model (HMM). The recognition system showed 99.01% recognition rate on the proposed language set. This result validates that the proposed language set enhances discernaility among the hand motions such that the performance of hand signal recognizer is improved.

Full Azimuth Multiple Sound Source Localization with 3-Channel Microphone Array

SHON, Suwon,HAN, David K.,BEH, Jounghoon,KO, Hanseok The Institute of Electronics, Information and Comm 2012 IEICE transactions on fundamentals of electronics, Vol.95 No.4

<P>This paper describes a method for estimating Direction Of Arrival (DOA) of multiple sound sources in full azimuth with three microphones. Estimating DOA with paired microphone arrays creates imaginary sound sources because of time delay of arrival (TDOA) being identical between real and imaginary sources. Imaginary sound sources can create chronic problems in multiple Sound Source Localization (SSL), because they can be localized as real sound sources. Our proposed approach is based on the observation that each microphone array creates imaginary sound sources, but the DOA of imaginary sources may be different depending on the orientation of the paired microphone array. With the fact that a real source would always be localized in the same direction regardless of the array orientation, we can suppress the imaginary sound sources by minimum filtering based on Steered Response Power — Phase Transform (SRP-PHAT) method. A set of experiments conducted in a real noisy environment showed that the proposed method was accurate in localizing multiple sound sources.</P>