Audio Engineering Society Convention Paper 8079epubs.surrey.ac.uk/2937/2/2010 128th AES Convention...

Audio Engineering Society

Convention PaperPresented at the 128th Convention

2010 May 22–25 London, UK

The papers at this Convention have been selected on the basis of a submitted abstract and extended precis that have

been peer reviewed by at least two qualified anonymous reviewers. This convention paper has been reproduced from

the author’s advance manuscript, without editing, corrections, or consideration by the Review Board. The AES takes

no responsibility for the contents. Additional papers may be obtained by sending request and remittance to Audio

Engineering Society, 60 East 42nd

Street, New York, New York 10165-2520, USA; also see www.aes.org. All rights

reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct permission from the

Journal of the Audio Engineering Society.

Time and Level Localisation Curves For A

Regularly-Spaced Octagon Loudspeaker Array

Laurent S. R. Simon1 and Russell Mason1

1Institute of Sound Recording, University of Surrey, Guildford, Surrey, GU2 7XH, United Kingdom

Correspondence should be addressed to Laurent S. R. Simon ([email protected])

ABSTRACT

Multichannel microphone array designs often use the localisation curves that have been derived for 2-0stereophony. Previous studies showed that side and rear perception of phantom image locations requiresomewhat different curves. This paper describes an experiment conducted to determine localisation curvesusing an octagonal loudspeaker setup. Various signals with a range of interchannel time and level differenceswere produced between pairs of adjacent loudspeakers, and subjects were asked to evaluate the perceivedsound event’s direction and its locatedness. The results showed that the curves for the side pairs of adjacentloudspeakers are significantly different to the front and rear pairs. The resulting curves can be used to derivesuitable microphone techniques for this loudspeaker setup.

1. BACKGROUND

A number of studies have shown that localisationto the side and rear of a listener in a system withonly two rear loudspeakers - such as quadraphonic(denoted in this paper as 2-2) or ITU-R BSS.775-1[1], more generally known as the 5.1 surround sound(3-2) - is problematic. Theile found that the locali-sation and locatedness to the side of the listener on a60 ◦-spaced pair of loudspeakers is less precise thanto the front [2]. In addition, others have found thatlocalisation to the side in a 3-2 system is poor, e.g.

[3] and [4]. If the intention is to enable audio record-ings to reproduce sound sources around the full 360 ◦

reproduction, while still based on the summing lo-calisation principles, a different loudspeaker array istherefore required.

In [5], it was explained that full 360 ◦ reproductionin the horizontal plane requires a homogeneous sys-tem which has better localisation capabilities to theside and to the rear of the listener in comparison tothe 3-2 system. One approach to enable similar lo-calisation performance around the full 360 ◦ of the

8079

Simon AND Mason Localisation Curves for a Regularly-spaced Octagon Loudspeaker Array

45°

90°

135°

180°

Figure 1: The octagon loudspeaker setup used inthe experiment

horizontal plane is to use a loudspeaker setup whereeach pair of adjacent loudspeakers (referred to inthis paper as a segment) has the same subtendedangle. As the system had to remain simple, and fortechnical reasons have at most eight loudspeakers,an octagon configuration was chosen, as shown infig. 1. A previous experiment demonstrated thatthis system provides relatively good localisation andmore even locatedness around the 360 ◦ of the hori-zontal plane, compared to a 3-2 system, when usingVector-Based Amplitude Panning (VBAP) [6].

In order to develop microphone techniques for thisarray, it is useful to derive appropriate localisationcurves. These can be used to aid the design of ar-rays by predicting the perceived location of sourcesignals based on analysis of the relative level andtime differences between microphones.

Multichannel microphone array design is often basedon frontal stereophonic (stereo) localisation curves,such as those presented by Williams [7], Wittek[8] and Lee [9]. The majority of these localisationcurves have been created through subjective exper-imentation using stimuli reproduced over a conven-tional 2-channel (2-0) stereo configuration (in which

a pair of loudspeakers are positioned on the hori-zontal plane, symmetrically one either side of themedian plane). These have often then been appliedto developing surround sound multichannel micro-phone arrays where the loudspeakers are positionedaround the listener. In some cases, the localisationcurves have been adapted for the new loudspeakerconfiguration, in others they have been applied di-rectly.

Williams [10] applies the 2-0 localisation curves to allthe pairs of adjacent microphones in his microphonearrays, independently of the subtended angle of agiven loudspeaker pair and the position of the pairin relation to the listener. His hypothesis is thatthe localisation curves remain constant for all thesegments of a 3-2 system.

Theile [11] adapted the 2-0 localisation curves forthe front three channels of a 3-2 system by assum-ing that they are applicable as long as the phan-tom source (i.e. “the apparent location of the soundsource in-between loudspeakers” [12]) position isnot expressed in terms of angle in degrees but interms of angle shift in percentage. For a givenmicrophone and source signal configuration, if therecorded source signal is perceived at 10 ◦ off-centerright on a ± 30 ◦ loudspeaker setup (i.e. two-thirdsof the way across from one loudspeaker to the other),it will be perceived at 20 ◦ (again two-thirds) on a0 ◦ - 30 ◦ off-center right loudspeaker setup. Theiledoes not apply these curves to the use of the tworear channels, as he considers that these should onlybe used for surround effect.

In addition, Theile showed that the localisationcurves between a pair of loudspeakers was depen-dant on the angle shift of that pair [2]. In otherwords, the localisation curve resulting from a pair ofloudspeakers in the horizontal plane positioned sym-metrically around the median plane was different tothat of a pair of loudspeakers rotated around thelistener so that the subtended angle is the same butone loudspeaker is towards the front and the othertowards the rear. However, studies to determine lo-calisation curves have only been undertaken on asmall subset of possible loudspeaker arrangements,particularly when considering positions to the sideand to the rear of the listener; some examples in-clude Thiele and Plenge 1977 [2], Martin et al. 1999[3] and Kim et al. 2008 [4].

AES 128th Convention, London, UK, 2010 May 22–25

of 19


Based on this research, it is apparent that localisa-tion curves need to be determined for the 8-channelsystem as a tool to ease development of appropriatemicrophone arrays. In view of this, an experimentwas conducted to determine the localisation curvesfor each segment (i.e. each pairing of adjacent loud-speakers). Depending on the directivity of the mi-crophones selected and their spacing (if any), bothinterchannel level differences (ICLDs) and interchan-nel time differences (ICTDs) could result. The lo-calisation curves measured in this experiment aretherefore both time and level dependent.

The first section of this paper describes the experi-ment set-up used in both a pilot experiment and inthe main experiment. The pilot experiment, used toevaluate the method and select the most consistentlisteners, is described and the results displayed. Themain experiment is then described, and the resultsare discussed in comparison to those derived previ-ous for other loudspeaker layouts.

2. EXPERIMENTAL DESIGN

2.1. Selection of experimental conditions

In order to create localisation curves for each of thesegments in the loudspeaker array, stimuli with arange of interchannel level differences (ICLDs) andinterchannel time differences (ICTDs) were required.A positive ICLD between two loudspeakers A and Bmeans that the level of the signal emitted by theloudspeaker B is louder than the signal emitted bythe loudspeaker A. A positive ICTD between twoloudspeakers A and B means that the signal emittedby the loudspeaker B is delayed compared to thesignal emitted by the loudspeaker A.

The same sets of ICLDs and ICTDs were used forall of the loudspeaker segments, to allow for equalcoverage of the full 360 ◦ of azimuth, and to allowcomparison between the segments.

The range of ICLDs and ICTDs were chosen basedon previous research into perception of two-channelloudspeaker reproduction. According to Blauert[13], an ICLD of between 12 and 18 dB leads toa phantom source being perceived in one of theloudspeakers, and the ICTD that causes a phantomsource to be perceived in one of the loudspeakers is1.1 ms. However, an informal test showed that al-though this is true for a stereophonic setup, a larger

ICTD seemed to be necessary to the side of the lis-tener, and a maximum ICTD of 1.5 ms was thereforechosen. The ICLDs and ICTDs were therefore var-ied across this range in equal steps, sampling therange at intervals that were a compromise betweenresolution and practicality. ICLD varied thereforebetween -18 and +18 dB, in steps of 3.6 dB, whileICTD varied between -1.5 and +1.5 ms, in steps of0.3 ms.

Wittek showed that in the case where there is a com-bination of ICLD and ICTD, the phantom sourceshift (i.e. the angle between the middle of the loud-speaker segment and the perceived direction of thephantom source) is equal to the sum of the phantomsource shifts of the ICLD and ICTD [14]. However,this effect will be limited to the subtended angleof the loudspeakers, in that the summation of thephantom source shifts resulting from the ICLD andICTD will not cause the phantom source to movepast either of the loudspeakers reproducing the stim-ulus. Based on this, as a stimulus with an ICLD of18 dB is likely to be perceived as a phantom sourcelocated at the same place as the loudest loudspeaker,the addition of a negative ICTD (i.e. making thelouder loudspeaker relatively earlier in time) is un-likely to make the phantom source move further to-wards or past the louder loudspeaker. Likewise, astimulus with an ICTD of 1.5 ms is likely to beperceived as a phantom source located at the sameplace as the earlier loudspeaker, and the addition ofa negative ICLD (i.e. making the later loudspeakerrelatively quieter) is unlikely to make the phantomsource move further towards or past the earlier loud-speaker. In addition, based on Wittek’s work it wasalso expected that some combinations of intermedi-ate ICLD and ICTD values could lead to a phantomsource being located in a loudspeaker, and increasingeither of these values would not significantly changethe position of the phantom source. Hence, it wasfound unnecessary to test all of the possible combi-nations of ICTD and ICLD, and only intermediatevalues were combined, as shown in fig. 2.

It is also impossible, for a microphone array com-posed of microphones pointing outwards which havethe same directivity and equal spacing, to capture asound source with both a positive ICTD and a pos-itive ICLD (or both negative), as the microphonein which the sound arrives first will be the micro-


of 19


ICLD (dB)

ICTD (ms)

3.6 7.2 10.8 14.4 18-3.6-7.2-10.8-14.4-18

0.3

0.6

0.9

1.2

1.5

-0.3

-0.6

-0.9

-1.2

-1.5

Points of measurement of localisation and locatedness

Figure 2: Combinations of ICLD and ICTD thatlead to an evaluation of direction and locatedness oneach loudspeaker segment and for each sound source

phone that is the most directed towards the soundsource. For this reason, the combinations of ICTDsand ICTLDs mainly had differing polarities. How-ever, two further ICTD and ICLD combinations,each having the same polarity, were introduced (-3.6 dB, -0.3 ms and +3.6 dB, +0.3 ms) to cover thecase of an heterogeneous microphone array contain-ing a combination of supercardioid and omnidirec-tional microphones and that would have up to 90 ◦

between two adjacent microphone capsules.

This combination of ICLDs and ICTDs resulted in43 conditions for use in the experiment, as shown infig. 2. Each of these combinations were used for eachof the 8 segments of the loudspeaker array (only sig-nals involving adjacent pairs of loudspeakers - eachsegment - were used in this experiment), meaningthat there were 344 conditions in total.

The source signals used in the experiment were cho-sen to contain a range of temporal and spectralcharacteristics; they were pink noise, female speech,cello, and bongos. The bongo sound included manytransients, whereas the cello sound contained fewtransients. Transients are important in auditory

spatial perception as they are a strong cue for de-tecting Interaural Time Differences (ITD) [15], espe-cially at high frequencies where the interaural phaseof a signal cannot be detected due to the breakdownof phase locking in the ear [13]. Hence the pres-ence or absence of these cues in the bongo and cellosignals respectively could be used to evaluate theimportance of these on the results. The noise signalhad a wide frequency content, giving strong Inter-aural Level Difference (ILD) cues (one of the othermain localisation cues, used mostly in high frequen-cies [13]) and IPD cues (used mostly below 800 Hz[13]). A voice signal was also included because of thevariety of inherent cues: fricatives (noise-like), plo-sives (transient-like), and voiced sounds (more tonaland relatively continuous), offering a large variety oflocalisation cues.

If each source signal has been tested for each condi-tion, it would have resulted in 1376 stimuli to test.

In order to reduce the number of stimuli each listenerwould have to rate, it was assumed that the resultswould be symmetrical about the median plane, aswas found in the previous experiment [5]. Hence, thelisteners rated all of the source signals, and all of theICLD conditions, but for only half of the loudspeakersegments (i.e. only one side of the loudspeaker ar-ray). However, using only one side per listener mighthave introduced bias through repetition. Therefore,the side on which the stimuli were presented wasrandomised for each listener, with the stimulus pre-sentation arranged across pairs of listeners so thatall the conditions were rated an equal number oftimes and comparisons could be made between thetwo sides to verify the assumption of symmetry inthe results. A number of the stimuli were rated morethan once, in order to test the consistency of the lis-teners, leading to a total of 812 stimuli per listener.

2.2. Choice of perceptual attributes

The principal purpose of the experiment was to cre-ate localisation curves for use in designing micro-phone arrays. Therefore, the listeners were asked toindicate the perceived position of each stimulus asan angle around the horizontal plane.

In addition to the judgements of the stimulus loca-tion, the listeners were asked to rate the located-ness of the sound. Lund defined locatedness as ”thecertainty of a source’s localisation” [16] . This is


of 19


I am absolutely certain

I have a slight doubt

I have a doubt

I am really not sure

I have no idea

100

50

75

25

0

Locatedness scale

Figure 3: Locatedness scale

expected to be useful information for designing mi-crophone arrays, as depending on the intended ap-plication the sound engineer might want his micro-phone array to produce a very well localised phan-tom source or a phantom source whose location isnot certain.The listeners rated the locatedness on a scale of 0 to100, with labels each quarter of the scale, as follows:“I am absolutely certain of the phantom source’s po-sition”, “I have a slight doubt about the phantomsource’s position”, “I have doubts about the phan-tom source’s position”, “I am really not sure aboutthe phantom source’s position” and “I have no ideaof the phantom source’s position”. Fig. 3 showsthe scale used and how it relates to the locatednessvalues

2.3. Equipment and acoustic conditions

The experiment was conducted in a listening roomthat meets the acoustic specifications of ITU-RBS.1116 [17]. The loudspeakers were Genelec8020As, and these were placed on stands at approxi-mately ear height (1.35m), equally spaced 45 ◦ apart,1.5m from the listener, as shown in fig. 1. In order toreduce the influence of visual cues, the loudspeakerswere hidden behind a visually opaque and acousti-cally transparent curtain.To help listeners to determine the judged angle ofeach stimulus, a circular metal structure, 1 cm high

Figure 4: The user interface used for this experi-ment to indicate in which direction the listener per-ceived the phantom source to be, and how certainhe was about the phantom source’s direction.

and 2 m diameter, displayed the angles with 5 ◦ res-olution. The metal structure was placed 20 cm be-low loudspeaker level in order to reduce its influenceon the acoustic field. A user interface, designed byDewhirst [18], was provided that displayed the cur-tains, the listener’s head and similar angles to thoseindicated on the curtains. The perceived directionof each stimulus could then be indicated by the lis-tener by clicking on the user interface using a mouse,which displayed a pointer oriented in the chosen di-rection, as shown in fig. 4.

The stimuli were reproduced using a computer run-ning MaxMSP, which displayed the user interfaceand rating scales. The software randomised the or-der of presentation of the stimuli to reduce ordereffects. The stimuli were looped so that the lis-teners could take as long as they needed to makea judgement. For each stimulus, the listeners firstwere asked to indicate the location of the stimulus,and then were asked to rate the locatedness. Oncethis was done the software moved on to the nextstimulus.

The experiment was intended to allow derivation ofthe localisaton curves for the adjacent loudspeakerpairs all around the listener. If the listener had beenfree to move their head, then this would have af-fected the results (e.g. the listener may have endedup facing the active system segment each time,meaning that each segment would be in front when


of 19


considered from the listener’s point of view). Onthe other hand, it was considered that physically re-straining the listener’s head would have made judge-ment of the stimulus location more difficult, as it isdifficult to quantify an unseen position. Therefore,a system was introduced that allowed the listenersto move their head, but only reproduced the stimuliwhen they were facing forwards. To enable this, thelisteners wore a head tracker. If they moved theirhead by more than five degrees to any side fromdirectly in front, or by more than one inch in anyhorizontal direction, the sound would stop (using a30ms fade to remove distracting clicks). This en-abled them to move their head to check the angleswritten on the circular structure without being in-fluenced by the perception of the sound event whennot facing forward.

Mason and colleagues discussed the advantages anddrawback of different verbal and nonverbal elicita-tion techniques in the subjective assessment of thespatial attributes of an auditory event [19]. Theyconcluded that the most accurate elicitation meth-ods for localisation are egocentric-based methods,where the listener can point directly at the de-sired direction. However, such an elicitation tech-nique would be problematic in this experiment, asa method was necessary which disabled the stimu-lus when the listener moved. In this case, a listenerwould have difficulty using an egocentric pointingmethod as sounds to the rear would be difficult toindicate accurately without movement, and movingwould stop the stimulus reproduction and hence maycause errors due to inconsistent spatial references.In the article, Mason et al. also discuss the use oftwo-dimensional graphical representation of space.This raises the problem of translation of the ego-centric physical reference to a graphical reference.Mason et al. explain that this translation can bemade easier for the listener and errors can be re-duced by representing the listener and visual ob-jects around the listener on the user interface. Inthis experiment, the listener was represented on theinterface, as well as the surrounding angle markers.Whilst this method was potentially not as accurateas an egocentric method allowing free movement,this method was considered the optimum compro-mise given the limitations of the experiment.

3. PILOT EXPERIMENT

Initially, a pilot experiment was conducted to testthe experiment method and to select listeners. Forthis, a subset of the experiment stimuli was used,employing the method and setup described above.22 listeners took part in the pilot experiment, andthey rated 32 stimuli twice.The listener selection was predominantly based onthe intra-listener consistency: analysing the consis-tency of each listener across all the stimuli of thisexperiment. For each listener, a univariate analy-sis of variance (ANOVA) was carried out, where theICLD, ICTD, segment and source signal were en-tered as the independent variables, and either thejudged location or locatedness were entered as thedependent variables. The consistency of each lis-tener could then be judged from the mean squareerror term in the ANOVA results [20]. For each lis-tener, the square root was taken of each mean squareerror term (so the numbers were comparable to theoriginal scale), and then scaled to be a percentageof the whole scale. The scaled Root Mean SquareError (RMSE) was measured both when includingall cases and when including only cases where locat-edness was rated above 70%, meaning that in thelatter case, listeners thought they were certain ofthe sound source’s position, and could be expectedto be at the best of their consistency. Both methodsshowed similar results for most listeners, 10 of themhaving a scaled RMSE lower than 2% or close to 2%in both cases, as can be seen in the example shownin fig. 5. These 10 listeners were selected for themain experiment.The results of the pilot experiment were also used toverify the experimental method. A one-way ANOVAwas carried out to check the symmetry assump-tion: the folded-back judged location and located-ness were selected as dependant variables and theside on which the stimuli were reproduced was se-lected as the factor. Significance of the ANOVAfor the folded-back judged angle and for locatednesswere respectively 0.715 and 0.743. As they are bothabove 0.05, this means that the side on which thestimuli were reproduced was not a significant factorfor the perception of the phantom source’s directionnor for the phantom source’s locatedness.It was also checked that the variations in ICLD andICTD were perceived as expected - that is, a positive


of 19


0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 210

2

4

6

8

10

12

Subject number

Scal

ed R

oot M

ean

Squa

re E

rror

Scaled RMS error for the perceived angle, per subject

Figure 5: Scaled RMS error for each listener, com-puted on their evaluations of the judged angle. Thethreshold for selection was set to 2%. The scaledRMS error was evaluated in two different cases, andthe results shown here are the scaled RMS error mea-sured when including all cases.

ICLD led to a movement of the phantom source to-wards the loudest loudspeaker, that a positive ICTDled to a movement of the phantom source towardsthe loudspeaker emitting the earliest sound, and thatthe phantom source was always between, or close to,the loudspeakers that emitted sound.

However, a few front / back confusions were found.Front / back confusions were evaluated through anestimation of the amount of judged angle confusionfor each listener. For each stimulus, a listener wasconsidered as having confused the front / back posi-tion when the judged position was outside the sub-tended angle of the active pair of loudspeakers. Ascore of 0 was given to each listener for each stimulusperceived inside the loudspeaker segment that emit-ted sound. For each stimulus perceived outside ofthe loudspeaker segment that emitted sound, a scorecorresponding to the difference between the judgedangle and the angle of the closest active loudspeakerwas given. The mean of this out-of-segment scorewas then computed for each listener. Two subjectsout of the twenty-two were found to have an out-of-

segment confusion mean score that was significantlyhigher than the others, and were therefore excluded.As they were not part of the 10 listeners selectedbased on consistency, this did not influence the lis-tener selection.

Hence, it was found that the experimental methodproduced usable results, and the most consistent lis-teners were selected for the main experiment.

4. MAIN EXPERIMENT

The main experiment (using only the listeners se-lected in the pilot experiment), consisted of a largenumber of stimuli for each listener to rate. In orderto avoid tiredness, each listener undertook 7 sessionson different days, each session containing one famil-iarisation section of 10 stimuli and two sub-sessionsof 58 stimuli each. Listeners were allowed to havea break between each sub-session. The experimentemployed the method and setup described above,identical to the pilot experiment. All 10 selectedlisteners took part in the experiment. A subset of124 stimuli were rated twice to evaluate the listen-ers’ consistency. The remainder were rated once byeach listener.

4.1. Analysis

The localisation data resulting from the experimentwere judgements of the perceived azimuth as an an-gle on a scale of 0 ◦ to 360 ◦. In order to avoid scalediscontinuities and to convert the data onto a singlehemisphere (based on the assumption of left/rightsymmetry discussed above), translation was needed.Data was translated to a -180 ◦ to +180 ◦ scale. Thelocalisation data that corresponded to stimuli playedon the left hand side of the configuration were thenmapped onto the opposite hemisphere to representthe symmetry of the configuration. Finally, judgedangles between -180 ◦ and -90 ◦ were translated toangles between +180 ◦ and +270 ◦ to avoid scale dis-continuities from causing errors during the statisticalanalysis (e.g. the mean of +179 ◦ and -179 ◦ is 0 ◦

whereas the intended direction is likely to be 180 ◦.

The intra-listener consistency was analysed usingthe same technique as shown above, and it wasfound that for the location judgements, the scaledRMS error was similar to that of the pilot exper-iment. The consistency of the locatedness judge-ments was found to have a larger spread than the


of 19


pilot experiment. This was however thought to bedue to the larger number of stimuli and wider rangeof conditions under test. It was found that one lis-tener had rated 98% of the stimuli at the top of thescale and the remaining stimuli in the top 5% of thescale. These ratings differed from all of the otherlisteners’s ratings, whose ratings were normally dis-tributed on a range between approximately 75% and100%. This listener’s locatedness ratings were there-fore dismissed for the computation of locatednessdata.

In order to check that the data met the assump-tions of parametric statistical analysis methods, aKolmogorov-Smirnov test was carried out for eachexperimental condition. It showed that the vast ma-jority of the cases were normally distributed (80%of the localisation judgements and 85% of the locat-edness judgements). This means that in general theresults are suitable for parametric statistical analy-sis (such as ANOVA), but that non-parametric testsshould be considered in order to confirm results [ref-erence?].

A first repeated-measures ANOVA was carried outto check the assumption that the side on which thestimuli were reproduced was not a significant fac-tor: the folded-back judged angle was selected asthe dependant variable and the side on which thestimuli were reproduced, the combination of ICLD /ICTD (denoted later as Stimulus), the loudspeakersegment on which the audio was reproduced (Seg-ment) as well as the source signal (Signal) were se-lected as factors. The repeated-measures ANOVAfound that the Side and the interactions betweenthe Side factor and the others was non-significant inall cases (sig. > 0.05). This means that data can beused independently of the side on which the stimuliwere reproduced.

Another repeated-measures ANOVA was carried outon the data to evaluate the effect of the source sig-nal, loudspeaker segment and combinations of ICLDand ICTD on the judged angle and on the located-ness. A pre-test transformation was applied to thejudged angle to scale the results from each segmentto be similar: the judged angle was unaltered for allstimuli reproduced on the 0 ◦ to 45 ◦ loudspeaker seg-ment, reduced by 45 ◦ for stimuli reproduced on the45 ◦ to 90 ◦ loudspeaker segment, reduced by 90 ◦ forstimuli reproduced on the 90 ◦ to 135 ◦ loudspeaker

segment and at last, reduced by 135 ◦ for stimulireproduced on the 135 ◦ to 180 ◦ loudspeaker seg-ment. This prevented the judged angle from biasingthe significance of the loudspeaker segment on whichthe stimuli were reproduced.Mauchly’s test performed for the judged angleand for locatedness showed that the assumption ofsphericity was verified for the source signal (sig. =0.446 for the locatedness case and sig. = 0.137 forthe judged angle case) and for the Segment (sig.= 0.842 for the locatedness case and sig. = 0.054for the judged angle case). This means that therepeated-measures ANOVA results can be used as-suming sphericity [21].In the case of the judged angle the repeated-measures ANOVA test found Segment, Stimulus,Signal * Segment, Signal * Stimulus and Segment* stimulus were statistically significant (respectivelysig. = 0.000 and F = 3341.902, sig. = 0.000 and F= 262.850, sig. = 0.000 and F = 9.272, sig. = 0.000and F = 2.147 and sig. = 0.000 and F = 2.243), ascan be seen in Table 1. The other interactions werefound to be statistically insignificant. As a check,non-parametric Kruskal-Wallis tests were performedon this set of data, and they found that the type ofsignal used was not significant (sig. = 0.931) butthat both the stimulus and the loudspeaker segmentwere (sig. = 0.000).In the case of Locatedness, see table 2, the repeated-measures ANOVA test found Signal, Segment, Stim-ulus as well as the Segment * Stimulus interactionwere statistically significant (respectively sig. =0.007 and F = 5.114, sig. = 0.000 and F = 27.852,sig. = 0.000 and F = 3.956, sig. = 0.000 and F= 2.089). The other interactions were found to bestatistically insignificant. The Kruskall-Wallis testsperformed on this set of data showed that Signal,Segment and Stimulus were all significant factors(sig. = 0.000).To summarise, the combined repeated-measuresANOVA and Kruskal-Wallis results indicated thatit is necessary to examine the changes in judged lo-cation and locatedness results caused by ICLD andICTD (each pair of ICTD and ICLD leading to astimulus value) separately for each loudspeaker seg-ment, but that the source signal only caused a sta-tistically significant change in locatedness withoutany statistically significant interactions.


of 19


Tests of Within-Subjects Effects

Source Type III Sum ofSquares

df Mean Square F Sig.

Signal 572 3 191 1.312 0.291Error 3923 27 145Segment 16979004 3 5659668 3341.9 0.000Error 45726 27 1694Stimulus 1927064 42 45882 262.8 0.000Error 65983 378 175Signal * Segment 11347 9 1261 9.272 0.000Error 11014 81 136Signal * Stimulus 23813 126 189 2.147 0.000Error 99822 1134 88Segment * Stimulus 45592 126 362 2.243 0.000Error 182902 1134 161Signal * Segment * Stimulus 35690 378 94 1.058 .0.225Error 303736 3402 89

Table 1: Sphericity assumed results of repeated-measure ANOVA conducted on localisation

Tests of Within-Subjects Effects

Source Type III Sum ofSquares

df Mean Square F Sig.

Signal 6527 3 2176 5.114 0.007Error 10211 24 425Segment 86354 3 28785 27.85 0.000Error 24804 24 1034Stimulus 32699 42 779 3.956 0.000Error 66131 336 197Signal * Segment 2074 9 230 1.387 0.210Error 11961 72 166Signal * Stimulus 20603 126 164 0.981 0.542Error 167948 1008 167Segment * Stimulus 48893 126 388 2.089 0.000Error 187209 1134 186Signal * Segment * Stimulus 57038 378 151 1.054 .0.238Error 432744 3024 143

Table 2: Huynh-Feldt corrected results of repeated-measure ANOVA conducted on locatedness


of 19


ICLD (dB)

Perc

eive

d an

gle

(deg

rees

)

Localisation curves for segment 1as a function of ICLD

Figure 6: Localisation curves and 95% confidenceintervals for all source signals between loudspeak-ers positioned at 0 ◦ and 45 ◦ for different values ofICTD.

As for the pilot experiment, the results of the mainexperiment showed that there were a number of front/ back confusions. Front / back confusions in thelocalisation of an auditory event can be explained bythe symmetry of the head [13]. If these results wereincluded in the analysis, they could have a significantinfluence on both the means and the 95% confidenceintervals of the judgements of perceived angle. It wastherefore decided to remove the location judgementsthat were outside of the loudspeaker segment thatemitted sound by more than 30 degrees.

The means and associated 95% confidence intervalsof the judged location results caused by the changesin ICLD and ICTD are shown for each loudspeakersegment in fig. 6 to fig. 13.

As an alternative interpretation, a surface plot ofthe means of the location judgements caused by thechanges in ICLD and ICTD are shown for each loud-speaker segment in fig. 14 to fig. 17.

In figs. 6, 9, 10, 13, 14 and 17, it can be seen thatchanges in ICLD and ICTD cause the judged loca-tion in the segment between the 0 ◦ and 45 ◦ loud-

ICLD (dB)

Perc

eive

d an

gle

(deg

rees

)


Figure 7: Localisation curves and 95% confidenceintervals for all source signals between loudspeak-ers positioned at 45 ◦ and 90 ◦ for different values ofICTD.

speakers and in the segment between the 135 ◦ and180 ◦ loudspeakers to follow a monotonic and rela-tively smooth trend. In addition, the combinationof ICLD and ICTD values appears to result in arelatively linear addition of the judged location an-gle: the equiangle curves, i.e. the curves showing allthe pairs of ICLD / ICTD leading to a same judgedangle, are parallel to the y = x axis. This meansthat the angle shift of the phantom source is regularacross both ICLD and ICTD variations.

On the contrary, figs. 7, 8, 11, 12, 15 and 16 showthat, compared to the relatively smooth trends ofthe front and rear segments, for the side segmentsa smaller absolute value of ICLD is necessary forthe phantom source to be perceived close to one ofthe loudspeakers. This means that a small changein ICLD could result in a large change of perceivedposition. Also, the variations in ICTD up to an ab-solute value of 0.3 ms cause the judged position tochange a certain amount, but beyond this there islittle change in judged position caused by increas-ing the ICTD. Finally, the localisation maps for the


of 19


ICLD (dB)

Perc

eive

d an

gle

(deg

rees

)


Figure 8: Localisation curves and 95% confidenceintervals for all source signals between loudspeakerspositioned at 90 ◦ and 135 ◦ for different values ofICTD.

side loudspeaker segments are not symmetrical, incontrast to the front and rear loudspeaker segments:the limitation of variation caused by the ICTD seemsto have more effect on the rear half of each of theside loudspeaker segments. It can be noted that thephantom sources created by varying the ICLD canbe successfully moved across the whole range fromone active loudspeaker to the other, but varying theICTD across the range of values tested only movesthe phantom source across a limited range of posi-tions.

Locatedness was mostly rated in the top section ofthe scale, higher than “I have a slight doubt aboutthe phantom source’s position”. As expected, rearloudspeaker segments were rated lower than frontalloudspeaker segments, see figs. 18 and 21. This ispossibly due to the method of reporting the per-ceived location, as listeners cannot see behind themtherefore making position judgements more difficult.They were allowed to move their head, but the soundwas then faded out until they returned their headto the forward direction. Some of the listeners ex-

ICLD (dB)

Perc

eive

d an

gle

(deg

rees

)


Figure 9: Localisation curves and 95% confidenceintervals for all source signals between loudspeakerspositioned at 135 ◦ and 180 ◦ for different values ofICTD.

plained that because they could not rate the posi-tion while looking at the angles, they felt they hadto rate locatedness lower. On those loudspeaker seg-ments, the combinations of ICLD / ICTD leading toa phantom source being perceived around 135 ◦ ledto the worst locatedness ratings, especially for highvalues of absolute ICTD.

Locatedness was also rated lower for the cello thanfor the other source signals (see fig. 22). This isdifferent from the results obtained in [5], where ina similar a similar experiment which only involvedvariations in ICLD, noise was rated lower than theother source signals. The listeners had explained thenoise sometimes seemed to come from two distinctplaces, but did not report such a problem during thecurrent experiment.

It may be expected that there would be a corre-lation between locatedness and the variance of thejudged angle, as poor locatedness may be related toa difficulty in locating the phantom source, whichmay in turn result in greater variance in the judge-ments made by the listeners. This was examined


of 19


5 5 5 5

5

35

45

ICTD (ms)

Perc

eive

d an

gle

(deg

rees

)

Localisation curves for segment 1as a function of ICTD

Figure 10: Localisation curves and 95% confidenceintervals for all source signals between loudspeak-ers positioned at 0 ◦ and 45 ◦ for different values ofICLD.

by conducting a Pearson’s correlation coefficient forthe mean of the locatedness versus the variance ofthe judged angles. The test was found highly signif-icant (sig. = 0.000), with a Pearson’s r coefficientof -0.798. Fig. 23 shows the scatterplot of the per-ceived angle standard deviation versus locatednessmean. It shows a good correlation between the vari-ables, thus supporting our hypothesis.

5. DISCUSSION

The experiment results showed that the ICTD hadless influence on the judged angle to the side of thelistener than to the front, whereas the ICLD causeda larger variation in judged angle for a given ICLDcompared to the front. It is possible that the use ofa wider range of ICTD values may have caused thephantom sources to be judged at either of the activeloudspeakers, but the relatively small variation forabsolute values greater than 0.3 ms indicates thatthis is not necessarily the case.

Fig. 24 and fig. 25 show respectively the pureICLD and pure ICTD localisation curves measured

5 5 5 5

ICTD (ms)

Perc

eive

d an

gle

(deg

rees

)


Figure 11: Localisation curves and 95% confidenceintervals for all source signals between loudspeak-ers positioned at 45 ◦ and 90 ◦ for different values ofICLD.

in this experiment for each segment. It can be seenthat there is little difference between the localisa-tion curves measured to the rear of the listener andthose measure to the front of the listener, neitherin mean nor variance, despite the expectation thataccurate judgement of location would be more dif-ficult for stimuli at the rear. Both front and rearICLD localisation curves are linear in comparison tothe ICLD localisation curves measured to the sideof the listener. ICTD localisation curves show thatobtaining an angle shift large enough to localise aphantom source inside a loudspeaker is more diffi-cult to the side of the listener.

The results of these experiments were compared withMartin et al’s 1999 results [3] measured for two loud-speakers located at 0 ◦ and 30 ◦ and for loudspeakerslocated at 30 ◦ and 120 ◦. In order to compare theresults, Martin et al’s results were scaled to matchthe subtended angle between the loudspeakers usedin this experiment (based on Theile’s assumption ofscalability [11] discussed above). For example, anICLD causing a phantom source to be perceived at


of 19


5 5 5 5

ICTD (ms)

Perc

eive

d an

gle

(deg

rees

)


Figure 12: Localisation curves and 95% confidenceintervals for all source signals between loudspeakerspositioned at 90 ◦ and 135 ◦ for different values ofICLD.

30 ◦ on Martin et al’s frontal loudspeaker segment(i.e. in the right hand loudspeaker) is scaled in thesefigures to be 45 ◦.

Fig. 26 shows the difference between the perceivedangles Martin et al. measured in a case of pure ICLDand those measured in the current experiment forthe frontal and rear segments of the octagon. Itcan be seen that Martin et al’s curve and the frontand rear segment curves have a similar trend, al-though Martin et al’s perceived angles tend to becloser to the side loudspeaker. This might be dueto the fact that they were measured with the lateralloudspeaker at 30 ◦, which might require a smallerICLD to fully pan sources.

Fig. 27 shows the difference between the perceivedangles Martin et al measured in a case of pure ICTDand those measured in the current experiment forthe frontal and rear segments of the octagon. Onceagain, the curve measured by martin follow a trendsimilar to the curves measured to the front and tothe rear of the listener, but the phantom source tendto be perceived closer to the side loudspeaker in this

5 5 5 5

ICTD (ms)

Perc

eive

d an

gle

(deg

rees

)


Figure 13: Localisation curves and 95% confidenceintervals for all source signals between loudspeakerspositioned at 135 ◦ and 180 ◦ for different values ofICLD.

ICLD (dB)

ICT

D (

ms)

Localisation map between 0 and 45 degrees

perceived angle (degrees)

Figure 14: Localisation map for all source signalsbetween loudspeakers positioned at 0 ◦ and 45 ◦.


of 19


ICLD (dB)

ICT

D (

ms)




ICLD (dB)

ICT

D (

ms)




ICLD (dB)

ICT

D (

ms)




ICLD (dB)

ICT

D (

ms)

Locatedness map between 0 and 45 degrees

Locatedness rating

Figure 18: Locatedness curves for all source signalsbetween loudspeakers positioned at 0 ◦ and 45 ◦.


of 19


ICLD (dB)

ICT

D (

ms)


Locatedness rating


ICLD (dB)

ICT

D (

ms)


Locatedness rating


ICLD (dB)

ICT

D (

ms)


Locatedness rating


Source signal

VoiceBongosNoiseCello

95%

CI L

ocat

edne

ss

87.00

86.00

85.00

84.00

83.00

82.00

Locatedness as a function of source signal

Figure 22: Locatedness for each type of source sig-nal


of 19


20,0015,0010,005,000,00

95,00

90,00

85,00

80,00

75,00

70,00

Perceived angle standard deviation

Loca

tedn

ess m

ean

Correlation between the perceived anglestandard deviation and the locatedness mean

Figure 23: Scatterplot of perceived angle standarddeviation versus locatedness mean. The line corre-sponds to the linear fit of the curve.

Perc

eive

d an

gle

(deg

rees

)

ICLD (dB)

Comparison between the measured perceived anglesas a function of ICLD only, for each loudspeaker segment

Figure 24: Comparison between the localisationcurves measured in this experiment for ICLD vari-ation without any ICTD, for each loudspeaker seg-ment. Error bars show the 95% confidence interval.

5 5 5 5

Perc

eive

d an

gle

(deg

rees

)

ICTD (ms)

Comparison between the measured perceived anglesas a function of ICTD only, for each loudspeaker segment

Figure 25: Comparison between the localisationcurves measured in this experiment for ICTD vari-ation without any ICLD, for each loudspeaker seg-ment. Error bars show the 95% confidence interval.

Perc

eive

d an

gle

(deg

rees

)

ICLD (dB)

Comparison between Martin et al. and measured perceived anglesas a function of ICLD

Figure 26: Comparison between Martin et al’sICLD localisation curve between loudspeakers at 0 ◦

and 30 ◦, the perceived angles being scaled to 0 to45 ◦, and ICLD localisation curves measured duringthis experiment for the frontal and rear loudspeakersegments. Error bars show the 95% confidence in-terval.


of 19


5 5 5 5

Perc

eive

d an

gle

(deg

rees

)

ICTD (dB)

Comparison between Martin et al. and measured perceived anglesas a function of ICTD

Figure 27: Comparison between Martin et al’sICTD localisation curve between loudspeakers at 0 ◦

and 30 ◦, the perceived angles being scaled to 0 to45 ◦, and ICTD localisation curves measured duringthis experiment for the frontal and rear loudspeakersegments. Error bars show the 95% confidence in-terval.

case too.

Martin et al. also measured localisation curves tothe side of the listener, between loudspeakers locatedat 30 ◦ and 120 ◦. Fig. 28 and fig. ?? show thecomparison between the curves measured by Martinet al., scaled, and those measured on the octagonloudspeaker array for pure ICLD and pure ICTD.

It can be seen that the ICLD localisation curvesto the side of the listener show that the phantomsources tend to be attracted to the loudspeaker forthe three loudspeaker configurations. In compari-son, the ICTD localisation curves showed larger vari-ance in the results despite this not being reflectedin the locatedness ratings. It is possible that thiswas caused by difficulties in accurately indicatingthe perceived location of the sounds to the side, dueto the experimental method, or differences in thelocation perceived by each listener.

Finally, the results of this experiment were comparedwith the results of Kim et al [4], who determined lo-calisation curves for amplitude panning between twoloudspeakers located at 30 and 110 degrees. As for

Perc

eive

d an

gle

(deg

rees

)

ICLD (dB)

Comparison between Martin et al. and measured perceived anglesas a function of ICLD

Figure 28: Comparison between Martin et al’sICLD localisation curve between loudspeakers at 0 ◦

and 30 ◦, the perceived angles being scaled to 0 to45 ◦, and ICLD localisation curves measured duringthis experiment for the frontal and rear loudspeakersegments. Error bars show the 95% confidence in-terval.

the results above, Kim et al’s results were scaledto allow comparison with the data from this experi-ment.

Fig. 30 compares Kim et al’s localisation curve andthose measured in the current experiment. It can beseen that Kim’s curve has the same tendency as the45 to 90 ◦ localisation curve from this experiment.The position of the loudspeakers in Kim et al’s ex-periment was more similar to this loudspeaker seg-ment than to any other of the octagon configuration.However, Kim et al’s experiment did not evaluatethe ICLD necessary to fully pan a source signal.

6. CONCLUSION

An experiment was conducted to determine locali-sation and locatedness curves for an octagonal arrayof loudspeakers. It was found that the perceptionof a phantom source’s location and locatedness issymmetrical about the median plane on this config-uration. It was found that localisation curves varydepending on the specific loudspeaker segment, suchthat localisation curves derived from convention 2-0


of 19


5 5 5 5

5

35

Perc

eive

d an

gle

(deg

rees

)

ICLD (dB)

Comparison between Martin et al. and measured perceived anglesas a function of ICTD only

Figure 29: Comparison between Martin et al’sICTD localisation curve between loudspeakers at 0 ◦

and 30 ◦, the perceived angles being scaled to 0 to45 ◦, and ICTD localisation curves measured duringthis experiment for the frontal and rear loudspeakersegments. Error bars show the 95% confidence in-terval.

Perc

eive

d an

gle

(deg

rees

)

ICLD (dB)

Comparison between Kim et al. and measured perceived anglesas a function of ICLD

Figure 30: Comparison between Kim et al’s ICLDlocalisation curve between loudspeakers at 30 ◦ anda 110 ◦, the perceived angles being scaled to 0 to45 ◦, and localisation curves measured during thisexperiment. Error bars show the 95% confidenceinterval.

stereophony are not applicable to loudspeaker pairspositioned to the side. It was also found that thelocalisation curves are close to linear on the frontalsegments but that on the side segments, the ICTDhas limited effect whilst a small variation in ICLDcan lead to a large change in the phantom sourceposition.

The localisation curves were found to be symmet-rical around the middle of the loudspeaker segmentfor the front and rear segments (i.e. if a combina-tion of ICLD (α) and ICTD (β) lead to the phantomsource being perceived θ ◦ away from the middle ofthe loudspeaker segment, a combination -α and -βlead to the phantom source being perceived -θ ◦ awayfrom the middle of the loudspeaker segment).

The comparisons between the results of this exper-iment and the results of similar experiments con-ducted on different loudspeaker setups show thatwhen using a particular loudspeaker setup, it ispreferable to use localisation curves measured on thesame configuration of loudspeakers. However, in theabsence of such curves, the use of localisation curvesmeasured on a loudspeaker setup having small differ-ences of loudspeaker placement, scaled for the anglesof the loudspeaker setup in use, can be an acceptablecompromise, depending on the precision of localisa-tion required.

7. REFERENCES

[1] Lee, H. , “Recommendation ITU-R BS.775-1 -Multichannel stereophonic sound system withand without accompanying picture”, Interna-tional Telecommunication Union, 1992-1994.

[2] Theile, G. , Plenge, G. , “Localization of Lat-eral Phantom Sources”, Journal of the AudioEngineering Society, Vol. 25, issue 4, pp. 196-200, April 1977.

[3] Martin, G. , Woszczyk, W. , Corey, J. , Ques-nel, R. , “Sound Source Localization in a Five-Channel Surround Sound Reproduction Sys-tem”, presented at the AES 107th convention,New York, United States, 1999, September 24-27. Preprint 4994.

[4] Kim, S. , Ikeda, M. , Takahashi, A. , “An op-timized pair-wise constant power panning algo-rithm for stable lateral sound imagery in the


of 19


5.1 reproduction system”, presented at the AES125th convention, San Francisco, United States,2008, October 2-5. Preprint 7602.

[5] Simon, L., Mason R., Rumsey F., “LocalisationCurves for a Regularly-Spaced Octagon Loud-speaker Array”, presented and the AES 127thconvention, New York, United States, 2009, Oc-tober 9-12. Preprint 7915.

[6] Simon, L. , transfer report, Institute of SoundRecording, University of Surrey, United King-dom, 2010.

[7] Williams, M. , “Unified Theory of MicrophoneSystems for Stereophonic Sound Recording”presented at the AES 82nd convention, London,United Kingdom, 1987 March 10-13. Preprint2466.

[8] Wittek, H., “Untersuchungen zur richtungsab-bildung mit L-C-R hauptmikrofonen”, Master’sThesis, Institut fr Rundfunktechnik, Germany,2000

[9] Lee, H., Rumsey, F., “Elicitation and Gradingof Subjective Attributes of 2-Channel PhantomImages”, presented at the AES 116th conven-tion, Berlin, Germany, 2004 May 8-11. Preprint6142.

[10] Williams, M. , “Magic Arrays Multichan-nel Microphone Array Design Applied to Mi-crophone Arrays Generating Interformat Com-patability”, presented at the AES 122nd con-vention, Vienna, Austria, 2007 May 5-8.Preprint 7057.

[11] Theile, G. , “Natural 5.1 Music RecordingBased on Psychoacoustic Principals”, presentedat the AES 19th international conference, Ger-many, 2001 June 21-24.

[12] Rumsey, F., McCormick, T., “Sound andRecording: An Introduction”, Focal Press,ISBN: 0240519965, 2005.

[13] Blauert, J. , “Spatial Hearing - The Psy-chophysics of Human Sound Localization”,MIT Press, ISBN: 0262024136.

[14] Wittek, H., “The Recording Angle - Based onLocalisation Curves”, presented at the AES112th convention, Munich, Germany, 2002 May10-13. Preprint 5568.

[15] Henning, G., “Detectability of interaural delayin high-frequency complex waveforms”, Journalof the Acoustical Society of America, vol. 55,issue 1, pp. 84-90, January 1974.

[16] Lund, T., “Enhanced Localization In 5.1 Pro-duction”, presented at the AES 109th conven-tion, Los Angeles, United States, 2000, Septem-ber 22-25. Preprint 5243.

[17] “Recommendation BS.1116 : Methods forthe subjective assessment of small impair-ments in audio systems including multichannelsound systems”, International Telecommunica-tion Union, 1994-1997.

[18] Dewhirst, M., “Modelling perceived spatial at-tributes of reproduced sound”, Doctor of Phi-losophy’s Thesis, Institut of Sound Recording,University of Surrey, United Kingdom, 2009.

[19] Mason, R., Ford, N., Rumsey, F., de Bruyn, B.,“Verbal and Nonverbal Elicitation Techniquesin the Subjective Assessment of Spatial SoundReproduction”, Journal of the Audio Engineer-ing Society, Vol. 49, issue 5, pp. 366-384, May2001

[20] Rumsey, F. , “Subjective Assessment of theSpatial Attributes of Reproduced Sound”, pre-sented at the AES 15th International Confer-ence, Copenhagen, Denmark, 1998.

[21] Field, A., “Discovering Statistics Using SPSS”,SAGE Publications LTD, ISBN: 1847879071


of 19

Audio Engineering Society Convention Paper 8079epubs.surrey.ac.uk/2937/2/2010 128th AES Convention...

Documents

Transcript of Audio Engineering Society Convention Paper 8079epubs.surrey.ac.uk/2937/2/2010 128th AES Convention...