서지주요정보
Acoustic focusing using beamforming for augmented sensing = 감각 증강을 위한 빔형성 기법에 근거한 음향학적 집속
서명 / 저자 Acoustic focusing using beamforming for augmented sensing = 감각 증강을 위한 빔형성 기법에 근거한 음향학적 집속 / Peter Gormsen.
발행사항 [대전 : 한국과학기술원, 2016].
Online Access 원문보기 원문인쇄

소장정보

등록번호

8029964

소장위치/청구기호

학술문화관(문화관)B1층 보존서고

MME 16077

휴대폰 전송

도서상태

이용가능(대출불가)

사유안내

반납예정일

리뷰정보

초록정보

Selective listening is a key feature of the human auditory system. It is the ability to selectively listen to one specific talker or voice in presence of multiple talkers, voices and/or background noise. The auditory system does this by using complex binaural signal processing that can exploit directional, spectral and visual cues. The task of selective listening in such complex sound environments has been termed the "cocktail party problem". Studies have shown that listeners with sensorineural hearing loss experience greater difficulty in these situations. Much attention has been given to this issue by research institutions and hearing aid manufacturers. Recently, different solutions based on the use of microphone array beamforming have been proposed. These solutions can generally be divided into fixed beamforming or adaptive beamforming systems. Fixed beamforming systems do not depend on the input, whereas adaptive beamforming systems can adapt their directional properties depending on the sound field. Adaptive systems can potentially achieve greater performance in certain situations, but fixed beamforming systems are generally said to be more robust and versatile. In this thesis, a solution to the cocktail party problem based on fixed beamforming for a microphone array will be described. A special trait of the solution proposed is the focus on obtaining a frequency invariant response. Frequency invariant beamforming avoids a coloration of the sound that most conventional methods do not. Yet, existing research on the subject has been found to be very limited. To investigate the performance of the proposed method, simulations and measurements have been conducted. Ideally, subjective testing should be performed to test the speech intelligibility of subjects using the proposed system method to other available methods. Unfortunately, such a test is very extensive and beyond the scope of this thesis. Lastly, interesting topics worthy of further research will be discussed.

선택적듣기 능력은 인간 청각 기관의 핵심 기능이다. 이것은 여러 화자의 음성과 배경 소음이 동시에 존재할 때 선택적으로 특정 화자 음성을 들을 수 있는 능력을 말한다. 청각 시스템은 복잡한 양귀에 들리는 신호를 이용한 복잡한 신호처리를 통하여 방향성, 주파수 대역 정보를 얻고 이와 시각적 정보를 종합하여 이러한 능력을 발휘하는 것으로 알려져 있다. 이러한 매우 복잡한 음향 환경에서의 선택적 듣기 능력을 통상 "칵테일 파티 문제" 라고 부른다. 연구에 따르면 감각신경성 청력손실이 있는 청자의 경우, 이러한 환경에서 큰 어려움을 겪는 것으로 보인다. 않은 연구기관 특히 보청기 업체들이 이 문제에 많은 관심을 보이고 있다. 최근에 마이크로폰 어레이에 기반한 빔형성 기법을 응용하는 다양한 해법들이 제안되었다. 이러한 해법은 고정 빔형성 과 적응형 빔형성 기법으로 나눌 수 있다. 고정 빔형성 법은 입력에 대하여 빔형성 특성이 고정적이나, 적응형의 경우에는 음장의 변화에 따라 지향성을 바꾸어 줄 수 있다. 적응 형 시스템은 잠재적으로 특정한 상황에서 높은 성능을 달성할 수 있지만, 고정 빔 형성 시스템은 일반적으로 더 강건하고 다용도로 사용 가능한 것으로 알려져있다. 본 연구에서는 마이크로폰 어레이와 고정형 빔형성 법을 사용한 칵테일 파티 문제의 해결 방안을 제안하고자 한다. 주파수에 따른 빔형성 특성 변화가 최소화되도록 설계하는 것에 초점을 맞추어 해결 방안을 제시하고자 한다. 일반적인 빔 형성 기법은 주파수 영역에 따라 다른 특성을 보여 착색 효과가 현저할 수 있으나 주파수 불변 빔형성 기법은 이러한 착색효과를 최소화 내지는 방지 할 수 있다. 본 연구 주제에 대한 기존의 연구는 매우 제한적인 것으로 보인다. 제안된 방법의 성능을 모의 실험과 측정 실험을 통하여 알아 보았다. 이상적으로, 주관적인 테스트를 통하여 제안한 방법과 기존의 기법을 비교하여 음성 명료도를 측정하는 것이 필요하나, 이는 대규모의 주관 평가를 필요로 하여 본 연구의 범위를 넘어선다. 마지막으로, 가치있는 추가 연구 방향 내지는 주제를 제안한다.

서지기타정보

서지기타정보
청구기호 {MME 16077
형태사항 xii, 71 p. : 삽화 ; 30 cm
언어 영어
일반주기 저자명의 한글표기 : 피터 곪슨
지도교수의 영문표기 : Youngjin Park
지도교수의 한글표기 : 박영진
Including Appendix
학위논문 학위논문(석사) - 한국과학기술원 : 기계공학과,
서지주기 References : p. 69-71
QR CODE

책소개

전체보기

목차

전체보기

이 주제의 인기대출도서

Sketch of the device proposed to solve the cocktail party problem: a microphone array mounted on the head of the user.

General beamformer (left) and delay-and-sum beamformer (right) with M microphones. For the general beamformer, the signal from each microphone is filtered by a transfer function H,(w) before summation. For the delay-and-sum beamformer, a delay represented as Ti is added to each signal before summation.

Left: Sketch of the delay-and-sum beamformer. Tm de- scribes themicrophone array distribution; ki is the wavenumber vector; and X2 is the steering angle. Right: DSB response of a free-field planar microphone array. 5 microphones are arranged in a line with a spacing of3 cm. Theplotshows the amplitude of the response as a function of frequency and steering angle.

Response of the superdirective microphone array beam- former for 2D (left) and 3D (right) diffuse noise fields. The microphone array consists of 5 microphones with a spacing of3 cm and is placed in free-field. In this scheme, an uncorrelated noise field would yield a response equal to the delay-and-sum beamformer seen in Figure 2.2 (right).

Spherical coordinate system (Meyer, 2001). The spherical coordinate system with (r,8,0) as theradial distance,polarangle and az- imuth angle, respectively. The microphone array distribution is placed along the vector Tmr and a is the radius of the sphere. Since the micro- phones are placed on the surface of the sphere, a = r in the following derivations.

The sound pressure on a rigid sphere with radius of10 cm for frequencies in the range of250-8000 Hz. The source position (assum- ino in far-field) is0 deo horizontallv

Mode amplitude (magnitude ofCn coefficients) ofthe first four Fourier terms as a function ofkR for a circular array of omnidirec- tional microphones in free field (left) and mounted on the equator of a rigid sphere (right).

Mode amplitude (magnitude ofCn coefficients) at different frequencies as a function ofmode number fora circular array ofomnidi- rectional microphones in free field (left) and mounted on the equator of a rigid sphere (right).

Sketch of various microphonemounting methods (left) and DSB response ofvarious microphone array mounting methods onarigid sphere with radius 10 cm (right). In all scenarios, 6 microphones are used. In the case of front-mounting and side-mounting, microphone spacing is 3 cm.

DSB response using a front-mounted microphone array with a varying number of microphones. The microphone array has a length of 16 cm (independent of the number of microphones) and is mounted on the equator ofa rigid sphere ofradius 10 cm.

Mapping between the s-plane (left) and z-plane (right).

Direct-Form II representation of the IIR filter

Transposed Direct-Form II representation of the IIR filter

Various target functions defined by the delay-and-sum beamforming (DSB) response at specific frequencies. The response is calculated by simulating an array of7 microphones witha 3 cm spacing mounted on a rigid sphere with radius 10 cm. Only the mainlobe ofthe DSB response is kentandbelow -20dBisused as "accented reoion"

Filter coefficients (left) are found by optimizing the re- sponse ofthe microphone array, 6 microphones with a spacing of3 cm, to a target function. The response ofthe microphone array using the fil- ter coefficients found are shown together with the target function for 4 kHz (right).

Simulated FIB response based on single frequency trans- fer function optimization. Microphone array uses 2-9 microphones mounted with a spacing of 3 cm on the equator of a rigid with radius 10 cm. Target function frequency is 2 kHz as seen in Figure 3.1.

Examples of fitting an IIR filter to the obtained response seen in Figure 3.3. The examples shown are for microphone no 3 in a microphone array of 6 microphones total and 3 cm spacing. The target filter coefficients are obtained for one frequency at a time and the total response can be seen in Figure 3.3 with 6 microphones.

Total beamforming response using the fitted IIR filter CO- efficients. The microphone array used includes 6 microphones with a spacing of3 cm. The microphone array is mounted on a rigid sphere with radius 10 cm.

Simulated FIB response based on 6th order IIR filter opti- mization. Microphone array uses 2-9 microphones mounted witha spac- ingof3 cm on the equator ofa rigid with radius 10 cm. Target function frequencv is2 kHz as seen in Fioure 3.1

Simulated FIB response based on 6th order IIR filter opti- mization. Microphone array uses 9 microphones mounted with a spac- ing of 3 cm on the equator of a rigid with radius 10 cm. Target func- tion frequencies are 1.2, 2, 3 and 4 kHz (top-left, top-right, bottom-left, bottom-right) as seen in Figure 3.1.

Sketch of measurement setup used for measuring KEMAR transfer function. The KEMAR head is placed on a turntable SO the an- gles can be recorded exactly. The microphone outputs are inserted in the Pulse system controlled by a computer. White noise was generated by an external device (Pulse can also be used as a generator) and the output was directed to both Pulse (for comparison) and through an amp

FIB response based on 6th order IIR filter optimization of measured KEMAR transfer functions. Microphone array uses 2-9 micro- phones mounted at nose level. Spacing is 6 cm for 2 to 5 microphones and 3 cm for 6 to 9 microphones. Target function frequency is 2 kHz as ceen in Fionire 31

FIB response based on 6th order IIR filter optimization of measured KEMAR transfer functions. Microphone array uses 9 micro- phones mounted atnose level with a spacing of3 cm. Targetfunction fre- quencies are 1.2,2,3 and 4kHz (top-left, top-right, bottom-left, bottom- right) as seen in Figure 3.1.

The developed wearable electret microphone array device mounted on the KEMAR. Array consists of7 microphones with a spac- ingof3 cm.

FIB response based on 6th order IIR filter optimization of microphone array device mounted on KEMAR. Microphone array uses 2-7 microphones mounted atnose level (see Figure 4.4). Spacingis6 cm for 2 to 4 microphones and 3 cm for 5 to 7 microphones. Target function frequency is 2 kHz as seen in Figure3.1.

FIB response based on 6th order IIR filter optimization of microphone array device mounted on KEMAR (see Figure 4.4). Micro- phone array uses 7 microphones mounted at nose level with a spacing of3 cm. Target function frequencies are 1.2, 2, 3 and 4 kHz (top-left, top-right, bottom-left, bottom-right) as seen in Figure 3.1.

Simulated FIB response based on 6th order IIR filter op- timization. Microphone array uses 2-9 microphones mounted on the equator of a rigid sphere with a spacing of 3 cm. Target function fre- OUCCIVES10101 30 ceen in mionire 31

Simulated FIB response based on 6th order IIR filter op- timization. Microphone array uses 2-9 microphones mounted on the equator of a rigid sphere with a spacing of 3 cm. Target function fre- auencv is3 kHz as seen in Fieure 3.1.

Simulated FIB response based on 6th order IIR filter op- timization. Microphone array uses 2-9 microphones mounted on the equator of a rigid sphere with a spacing of 3 cm. Target function fre- J166023. 고득에는 as ceen in Kionire 31

Measured normalized transfer-functions ofKEMAR.Spac- ingbetween themicrophonesis3 cm. Position 1 denotes themicrophone position above ear and position 5 is directly in the middle of the fore- head. Symmetry is used to calculate the transfer functions of the other side ofthe head.

Measured normalized transfer-functions of the in-ear mi- crophones on KEMAR.

FIB response based on 6th order IIR filter optimization of measured KEMAR transfer functions. Microphone array uses 2-9 micro- phones mounted at nose level. Spacing is 6 cm for 2 to 5 microphones and 3 cm for 6 to 9 microphones. Target function frequencyis 1.2kHz as ceen in Fionire 31

FIB response based on 6th order IIR filter optimization of measured KEMAR transfer functions. Microphone array uses2-9 micro- phones mounted at nose level. Spacing is 6 cm for 2 to 5 microphones and 3 cm for 6 to 9 microphones. Target function frequency is 3 kHz as seen in Figure3.1.

FIB response based on 6th order IIR filter optimization of measured KEMAR transfer functions. Microphone array uses2-9 micro- phones mounted at nose level. Spacing is 6 cm for 2 to 5 microphones and 3 cm for 6 to 9 microphones. Target function frequency is 4kHz as seen in Fioure 3.1

Measured normalized transfer-functions ofthe electretmi- crophone device on KEMAR head. Spacing between the microphones is 3 cm. Position 1 denotes the microphone position closest to the left ear and position 4is in themiddle ofthe forehead (see Figure 4.4).

Schematics ofthe pre-amplifier circuit used for the electret microphones. The operational amplifier used was National Semicon- ductor LM358N and the electret microphone used was Monacor MCE- 4000.

Response of the operational amplifier circuit depending on the value of the feedback resister (R2 in Figure C.2).

FIB response based on 6th order IIR filter optimization of microphone array device mounted on KEMAR. Microphone array uses 2-7 microphones mounted at nose level (see Figure 4.4). Spacing 1S 6 cm for 2 to 4 microphones and 3 cm for 5 to 7 microphones. Target function frequency is 1.2 kHz as seen in Figure3.1.

FIB response based on 6th order IIR filter optimization of microphone array device mounted on KEMAR. Microphone array uses 2-7 microphones mounted atnose level (see Figure 4.4). Spacingis 6 cm for 2 to 4microphones and 3 cm for5 to 7 microphones. Target function frequency is 3 kHz as seen in Figure3.1.

FIB response based on 6th order IIR filter optimization of microphone array device mounted on KEMAR. Microphone array uses 2-7 microphones mounted at nose level (see Figure 4.4). Spacing 1S 6 cm for 2 to 4 microphones and 3 cm for 5 to 7 microphones. Target function frequency is 4kHz as seen in Figure3.1.