Page  1 ï~~A STUDY INTO THE PERCEPTION OF ENVELOPMENT IN ELECTROACOUSTIC MUSIC Sarah Adair, Michael Alcorn, Chris Corrigan School of Music and Sonic Arts, Queen's University, Belfast ABSTRACT The majority of research into 'spatial impression' is focused on work in the area of concert hall acoustics. Such work explores two separate elements, namely 'apparent source width' (ASW) and 'listener envelopment' (LEV). Research in spatial impression within electracoustic music has described electroacoustic music as having an "expanded...spatial palette" that has the ability to create "complex interactions between spatial hearing and other domains of perception and cognition" (Kendall 2007). Taking this increased spatial complexity into consideration, this paper outlines an experiment which required individual participants to find the point of perceived envelopment within contrasting excerpts of electroacoustic music. The results of the experiment are presented and discussed in this paper. We conclude that, while envelopment in this music genre is principally different to concert hall acoustics, a comparable perception of envelopment was achieved by the participants. 1. INTRODUCTION 'Listener envelopment' (LEV) has been discussed extensively in the domain of concert hall acoustics. Marshall and Barron [2] differentiate between early lateral reflections that cause the source broadening associated with apparent source width (ASW), and LEV caused by late reverberant sound from the side resulting in the feeling of 'immersion'. Research by Morimoto and Maekawa (1989; cited in Morimoto et al, 2001) [3] describes LEV as the "degree of fullness of sound images around the listener", whilst Wakuda et al [4] define it as "the listener's sensation when the surrounding space is filled with sound images other than a sound image composing ASW" (Wakuda, 2003). Recent research (Morimoto [3]) has focused on reflections from above and behind the listener. The 'sense of envelopment' is a phrase that people often use when analysing an electroacoustic work. However as pointed out by Rumsey [5], in the domain of multi-channel reproduced sound, envelopment in the sense of concert hall LEV does not necessarily apply. In reproduced sound, envelopment can be described as either 'environmental' or 'ensemble'. Rumsey points out that "...subjects often use the term envelopment when they are surrounded by a number of dry sources in surround sound reproduction...this sensation is almost certainly not a property of late reflected sound, as the sources in question can be dry and direct, so it cannot be considered to be LEV in the traditional sense." (Rumsey, 2002) In this study the listener is surrounded by a number of sources that do not necessarily contain late reverberant energy, therefore we refer to envelopment in the sense of 'ensemble' envelopment. 2. METHOD In concert hall acoustics the frequency content and the amplitude and angle of reflections have been shown to have an impact on the perception of envelopment. The sense of envelopment is a subjective response that may be informed by issues of context and location within the space. This study focuses on understanding the responses of a number of participants in relation to 'envelopment' specifically within electroacoustic music. A simple experiment was devised to measure how subjects controlled the relative loudspeaker reproduction levels in a multi-channel loudspeaker configuration to achieve a sense of envelopment. Such a study might reveal significant results if similar levels would be chosen by each participant and if their decisions would be influenced by the spectral and gestural characteristics of the musical examples. 2.1. Experiment Space and Equipment The experiment was carried out using a subset of the 40 -loudspeaker system at the Sonic Arts Research Centre (SARC), Queen's University Belfast. Using a variation of Harrison's 'main eight' arrangement [6], subjects controlled the ground floor speaker levels using pots on a KillamixTM Mini USB MIDI controller. These pots are infinitely variable preventing participants from making visual judgments about reproduction levels. See figure 1 for a diagram of the speaker layout. Midi data from the controller was used to adjust the output levels of the sound files via a Max/MSP patch. Before the experiment began the loudspeakers were calibrated to reproduce identical levels at the listening position (70dBSPL, _0.5dB) from a continuous white noise source. All of the available acoustic wall panels in the Sonic Lab were employed resulting in a reverberation time of approximately 0.4 seconds at 1kHz. All participants were positioned in the same location during the experiment.

Page  2 ï~~2.2. Participants Thirteen people participated, one female and twelve males. The majority were in their late teens and 20s. All participants were students from SARC and all were F t d d Xii d ' R famlength ar with electro used acush with unique spectral and/or. PFrotfixedt oato gestural characteristics. The excerpts chosen were as follows" Pentes by Denis Smalley, chosen as it contains very continuous sounds of a granular nature; Cobra by Wayne Siegel, chosen for its use of a recognizable instrumental sound; Portrait d'un Visiteur by Christian Calon, chosen for its use of environmental sounds in the form of ocean waves; and Minuit by Christian Calon, chosen as it contains a high level of stereo interplay in Figucomparison to the other excerpts.layout 2.4. MusExperiment Procedure Four excerpts of music of between 1.30 - 2.30 minutes in length were used, each with unique spectral and/or The excerpts wstural characteristics.e replayed from the 'Front' loudspeakere as pair at a fixed reproduction level: a subjective "omfortable" list by Denig s Smalley, chovel wasen as selectd, whichntains equalled a midi value of 105, graiving the participantsy Waynscope to set the, controllable loudspeakers ('wide', 'side' instrumental sound; Portrait d'un Visiteur by Christian and 'rear') to a higher level if environmental sounds ing toe iformatiof ocn waves;from Cycling '74 regarding the 'gain-', object, steps of 10 equal ahigh level of stereo intferplayence ofin approximately 6dB. In the patch used for thiss. 2.4.experiment a value of 105 equates to a level of betweenedure 0 and +6dB. The aim of this study however was to establish a degree a correlation using midi values, not their equivalent signal levels. Prior to the experiment the participants were provideplayed wifromth a descriptiont' loudspeak ter 'envelopmenpair at a fixed reproduction level: a subjectiven the "comfortable" listening level was selected, which equalled a midi value of 105, giving the participants opportunity to listen to the excerpts before any data was recorded. Four participants chose to do this. The excerpts were presented in the order listed in 2.3 and the controllable loudspeaker levels were reset to the 0 midi value between each excerpt. At the end of the fourth excerpt the participants were asked to repeat the experiment, following which comments were gathered. 3. RESULTS Following the completion of the experiment, the data was analysed in terms of the responses to each excerpt, the response to each pair of loudspeakers and the responses by each individual participant. This analysis was accomplished through the creation of a number of data sets. The data set for excerpt information consisted of 13x2x3 results; that for speaker pair information consisted of 13x2x4 results; and that for participant information consisted of 2x3x4 results. The results were plotted in graphs and histograms, and statistical processes such as Standard Deviation, using the 'STDEVP' function in Excel, were applied in order to refine the results. Through these processes a number of conclusions have been made and are outlined in detail below. 3.1. Results by excerpt Despite the presence of a number of random results, there was definite correlation present in the graphs of the Siegel and two Calon excerpts, with the Siegel excerpt showing the strongest clustering of data. Carrying out statistical procedures we observed preference towards a specific range of levels, which we have called the 'envelopment range'. In both Calon excerpts and the Siegel excerpt this range was roughly between 90 and 120, with reference to midi values. Figure 2 presents data from the Siegel extract. Pentes by Smalley however appeared to show very little correlation. There was clustering in places, between 80 and 100; however correlation was weak and the majority of results appeared to be widely distributed across the entire range of values available to the participants............................................................................................................................................................................................................................................ Siegel Result Distribution 0 10 20 30 40 0 70 Resoi Figure 2. Distribution of results for the Siegel excerpt. 3.2. Results by speaker pair Inspecting each loudspeaker pair, we observed strong correlations in the side and rear speaker results. The majority of results in the side speakers fell within 90 and 110 and within 90 and 120 in the rear speakers. The

Page  3 ï~~wide speakers had the weakest correlation, with results widely spread across all choices of speaker level. As expected, when observing data for speaker pairs within each excerpt we saw similar correlations. The strongest correlation was present in the rear and side speakers respectively. With the exception of the Siegel excerpt the wide speakers showed the least correlation. Throughout the results for the Smalley and two Calon extracts we did see preference towards levels between 70 and 100. Despite this it is reasonable to say that many of the results were incomparable and spread across the entire range available. As stated however, the Siegel excerpt displayed a strong degree of correlation. With a few exceptions low in the level scale, all results were contained in the categories above 70, with the majority of results contained in the category labelled 90 to 99. 3.3. Results by participant Analysis of the differences between the first and second attempt of every participant revealed that 58% of second attempt results were within 10% of their first result. When extended to within 20%, this increased to 78%. Examining the participants in detail, we observed various similarities across all pieces. Four of the participants matched 100% of their second attempts within 20% of the first attempt, with a fifth participant achieving 91%. In comparison, one person matched 25% of their second attempts within 20% of the first attempt, with two further participants achieving 58%. Both of these participants did however perform well in the Siegel excerpt. Inspecting the average percentage changes for each piece within the context of speaker pairs, we observed that the largest changes in both groups were in response to the wide speakers. Looking at the overall response to each excerpt, the second Calon and Siegel excerpts had the lowest percentage change results. Supporting earlier trends, the Smalley excerpt had the highest overall average percentage change. The degree of variety in the responses was also observed when comparing average difference results. For the first group the average difference over all excerpts and speaker pairs was 26, whilst for the second group it was 77. Inspecting the participants of these groups in detail, we detected individual and general trends. For instance, despite participant number four performing 'well' in terms of percentage change, many of his/her results were of a very high level and outside the 'envelopment range'. With some exceptions many results belonging to the first group fell within a small range over all excerpts, attempts and speakers and within the 'envelopment range'. The side and rear speaker results were relatively constant within a small scale whilst the wide speakers produced the most changeable results. With a few exceptions the results of the second group covered a broad range of values and appeared random. In the case of participant number eleven there were large differences between attempts and little similarity between excerpts. There appeared to be few similarities in the wide and side speaker pairs in particular. 4. DISCUSSION 4.1. Observations Participants adopted various approaches to finding the envelopment level. In many cases turning their head and closing their eyes aided concentration on each speaker pair. Many felt that the rear and side speakers were the most important for envelopment so adjusted these levels first. Frequency was an important factor for some; however, one person commented that decisions were made based on high frequency content, whilst low frequency content had a greater impact on another. 4.2. Listening skills and aural awareness The experiment was repeated to establish the degree of correlation between first and second attempts from each participant. To avoid the perception that the first attempt was a practice, participants were not told that the test would be repeated. As the excerpts were played in sequence before their repetition, it was deemed unlikely that participants would be able to simply replicate identical responses within each excerpt. This indicates that we can trust the integrity of the second attempt results in this respect. Two approaches were taken by the participants to adjust the loudspeaker levels. The first approach involved a simple, systematic method of altering each loudspeaker pair in turn whilst listening carefully before fine-tuning their choice. In the case of participant thirteen, a process of adjusting the controls to find the envelopment range limits before choosing a position in the middle of the range was followed. The second approach appeared to involve quick decisions and little thought, as was evident in the results of participant eleven. Observations on the methods used to determine the point of envelopment and the level differences between each attempt highlight the listening skills and aural awareness of the individuals. In the majority of cases those who took time to listen seemed to produce similar results in both attempts. 4.3. Correlations between the results and excerpts Following the experiment, some participants commented on their difficulties with particular excerpts. Some found that excerpts containing substantial dynamic change and stereo interplay were distracting and difficult to judge, whilst others stated that this made it easier. For example, one person commented that the "more pronounced notes" of the Siegel excerpt were easier than the "long sustained sounds" in the Smalley excerpt. However in one case it was commented that the Siegel excerpt was difficult in terms of the rear speakers. This may be due to the fact that the excerpt consisted of the sound of a 'real' instrument and in a traditional concert hall setting the piano would not normally be located behind the audience. As previously stated, the Smalley excerpt showed the least correlation. This may be because it was the first

Page  4 ï~~excerpt of the experiment when participants were perhaps uncertain of what they were doing. Comparing the first and second attempts however we saw little change in the results. Further testing would need to be undertaken to establish if the weak correlation was due to the excerpt or the test procedure. With reference to the test procedure, two aspects that could be addressed in further testing, is randomisation of the excerpt presentation and the inclusion of a familiarisation session for all participants. Examining the results of the two groups, we saw little difference in the average percentage change between each piece for the first group, but a broader range of high results for both the excerpts and participants for the second group. The results revealed that participants in this second group had little agreement on the envelopment range for the Smalley piece in comparison to the Siegel piece. We could conclude that the nature of the excerpt had minimal influence on those people with good listening skills; however, this should be studied further. 4.4. Correlations between the results and speaker pairs The majority of envelopment tests have been conducted in the context of concert hall acoustics, therefore the decision was made to have a fixed level source at the front of the hall. It was also deemed important that participants had the same point of reference. The wide speakers showed the weakest correlation and the rear speakers the strongest. From comments we discovered that participants found the rear speakers difficult to hear at times, as they felt the front speakers were too loud. They felt the wide speakers were therefore unnecessary. In reality, the controllable speakers were capable of producing higher levels than the fixed speakers but few people took advantage of this. In a concert hall situation, the rear sound would never equal or be louder than sound from the stage; however, in this experiment the majority of participants preferred balanced sound from all directions. These comments highlight the differences between traditional LEV and envelopment in reproduced sound. 5. CONCLUSIONS From this initial study into perceived envelopment in electroacoustic music, we found strong correlations amongst the levels of envelopment that participants chose for each speaker pair and piece. We discovered that unlike concert hall acoustics, a high level of sound that does not necessarily consist of late reflections is required from the rear. In a recent study by Sazdov [7] an attribute labelled 'engulfment'' was put forth as a unique 3D spatial attribute. This study, which also took place in SARC, 1 Engulfment is a unique spatial attribute of 3D reproduced sound. It describes the sensation of being "'covered over' as opposed to simply being surrounded as defined by envelopment" (Sazdov, 2007). compared envelopment and engulfment using multichannel audio including elevated speakers. This study found that envelopment and engulfment were perceived as two independent but equally important spatial attributes. As well as investigating 'engulfment' as a spatial attribute, future work on the subject of envelopment would involve establishing the ratio of reproduced sound levels of the wide, side and rear loudspeakers relative to the main pair of speakers. Despite the lack of a standardised diffusion system, electroacoustic music performance is unique in comparison to traditional instrumental performance. As new diffusion systems are implemented and new dimensions are added, such as the system in SARC that incorporates speakers above and below an audience, we may experience new and unique spatial attributes that should be researched independently from concert hall acoustics. 6. ACKNOWLEDGEMENTS Thank you to all the students from SARC who gave up their time to take part and Ms Una Monaghan for her additional technical assistance when setting up the experiment. 7. REFERENCES [1] Kendall, G.S. "The artistic play of spatial organization: spatial attributes, scene analysis and auditory spatial schemata", ICMC 2007 Proceedings, Copenhagen [2] Marshall, A.H., Barron, M. "Spatial responsiveness in concert halls and the origins of spatial impressions", Applied Acoustics, 62, 2001, 91-108 [3] Morimoto, M., lida, K., Sakagami, K. "The role of reflections from behind the listener in spatial impression", Applied Acoustics, 62, 2001, 109-124 [4] Wakuda, A., Furuya, H., Fujimoto, K., Isogai, K., Anai, K. "Effects of arrival direction of late sound on listener envelopment", Acoustical Science and Technology, 24, 4, 2003. [5] Rumsey, F. "Spatial Quality Evaluation for Reproduced Sound: Terminology, Meaning and a SceneBased Paradigm", Journal of the Audio Engineering Society, Vol. 50, No. 9, 2002 [6] Harrison, J. "Sound, space, sculpture: some thoughts on the 'what', 'how' and 'why' of sound diffusion", Organised Sound 3(2): 117-27 [7] Sazdov, R., Paine, G., Stevens, K. "Perceptual Investigation into envelopment, spatial clarity, and engulfment in reproduced multi-channel audio", AES 31" International Conference, 2007