Ed: The Rockport Chamber Music Festival in Rockport, MA, is set to open its second season in its sumptuous Shalin Liu Performance Center on June 9, and concerts will run through July 17. In honor of the first anniversary of the dedication of Rockport Music’s new home, BMInt is pleased to publish a very interesting analysis of the Center’s acoustics by one of our writers, the eminent acoustician, David Griesinger. (Though many of the excellent concerts are already sold out, some tickets are likely still available.)
It is rarely possible to combine the pleasure of attending a great concert with the chance to do acoustic research, but just such an opportunity presented itself on Sunday March 13, at the occasion of an all-Brahms clarinet-piano concert in the new Shalin Liu Performance Center in Rockport MA. The performers were Eran Egozy, clarinet; Mike Hawley, piano, and Yukiko Ueno-Egozy, piano. The program contained the Intermezzi from Four Piano Pieces, op. 119, the Sonata for Clarinet and Piano in F minor, op. 120 No. 1, selections from the Liebeslieder Waltzes for four-hands piano, op 52a, and the Sonata for Clarinet and Piano in E flat Major, op. 120, No 2.
The performers all met for the first time in the music course at MIT given by David Deveau, the artistic director at Rockport Music, and all went on to dual careers as musicians and as top-flight scientists. Their principle interest has been in the emotional content of music – in the case of Eran Egozy, how emotion is generated in the process of playing, and how this creativity can be enabled through software such as Guitar Hero; and in the case of Mike Hawley, how the expression in a musical performance can be captured, coded, and reproduced.
So they jumped at the idea of using the concert as an opportunity to study the acoustics of the venue – and David Deveau was equally enthusiastic. I was interested because at my previous concert in the hall the Parthenia ensemble played with the glass window behind the stage covered by a woven wooden curtain. I found the sound in row G to be good, but marginal for the vocal soloist. I was very interested to hear how the hall worked acoustically when the glass was uncovered.
I have three microphone systems that record the sound pressure at a listener’s eardrums. Two are dummy heads with castings of my own pinna and ear canals. The third uses my own pinna and ear canals, equipping me with tiny microphones almost touching the eardrums. With this equipment I can effectively be in three places at the same time – and compare the sounds later.
I put one dummy in the center of row M on the floor, just in front of the overhanging balcony. This is the seat I occupied for the Parthenia. The other dummy was in the center of the fourth row in the balcony. I sat in row F on the floor for the first half of the concert, and in the center of the first row of the balcony for the second half. So we ended with recordings in four different seats which could be instantly switched back and forth. This allows the kind of close comparison of sound that not possible in a live concert. The recordings can also be compared by other individuals, so differences in preference can be explored. For technical measurements I set a Soundfield microphone (one that can identify the direction of sound) next to the dummy head in row G, and recorded three balloon pops from the stage just before the audience entered.
For me the experiment had a null result—the sound was fine in all four seats. Differences exist – and will be discussed later – but the sound of the piano and the clarinet during this concert was wonderful. The audience was delighted. They usually are. The music was great, the view through the window in front was pretty, and they had all paid good money to enjoy themselves. But science does not stop there. We would very much like to know why, even if the sound was successful. We need to find what is physically required for sound to be perceived as successful. And to do this we must confront the question of what constitutes “successful” sound.
It may be useful if we divide our perception of sound in a concert venue into three broad categories: loudness, clarity, and perceived reverberation. Loudness is probably perceptually the most important. Loudness in a concert hall increases with the number of the instruments playing and decreases with the number of people listening. But it also depends on hall design. If we design for high loudness, clarity will usually suffer. It is also easy to observe that the audibility of reverberation – or the sense of the hall – depends on loudness. It is easy to hear the hall when the orchestra plays loud, and quite difficult when they play soft. The perception of hall sound also depends on clarity – whatever clarity might mean. A listener cannot perceive the hall sound at all while the music is playing unless the sound of the instruments is distinctly heard as separate from the reflections and reverberation.
But perhaps the greatest problem is that there is no adequate definition of what clarity might mean or how to measure it. Unfortunately the standardized measure for clarity, C80, has only a limited relationship to our ability to hear either speech or music. Here are two examples of the same piano performance in two different venues (the meaning of LOC to be explained later): A dry piano recording was convolved with two measured binaural impulse responses to make these examples.
C80 = +5.0dB, (LOC = -3.4dB) link here
(This example is from row 5 in a small concert hall.)
C80 = +2.8dB, (LOC = +2.8dB) link here
(This example is from row R in a much larger hall.)
In spite of the lack of correlation with perception, C80 is routinely specified in hall design, and designs are modified until the specification is met. The result may be worse than if the measure was ignored.
Clarity for speech is often defined through rather crude tests of speech intelligibility. But speech is highly redundant; being able to understand most of what is said is a poor criterion for clarity. Recent work has shown that students in the rear of a poor classroom can understand the teacher, but using expectation, grammar, and context to make sense of the sound takes so much time and brain power that they cannot remember what was said. Students just a few rows closer do much better. We need a technical measure for “ease of remembering what was said.”
I am proposing that we solve this dilemma by defining clarity in a different way – through our ability to perceive separately or to understand one or more of several sounds occurring at the same time. For example, the ability to separate two people talking at the same time or to separate the direct sound from the reflections and reverberation that quickly obscure it.
With speech, the separation process is called the “cocktail party effect.” People with undamaged hearing have no trouble directing their attention to one of several simultaneous conversations, and if their name is mentioned in another conversation their attention involuntarily switches. But in the presence of reflections and reverberation, separating conversations becomes increasingly difficult. In poor conditions we are lucky to perceive only the loudest. (Think of trying to converse in a crowded restaurant.) I propose that the ability separately to perceive simultaneous sounds is the essence of clarity in music, classrooms, and concert stages. The ability to follow any one of several musical lines, or even several at the same time, is one of the joys of good acoustics.
The advantage of this definition is that we can use the physics of information to predict under what conditions source separation becomes impossible. Enough is known about the mechanics of our hearing organs to model the process our brain uses to decode the information in sound – and how this information is lost due to reflections and reverberation.
We know that source separation depends on pitch – and pitch perception in humans is circular in octaves. Humans can separately perceive the meaning of two monotone conversations if the pitches are different by half a semitone, or approximately three parts in a hundred. In my view, our amazing acuity to pitch evolved to make this separation possible. A musician can detect differences of pitch of about one part in one thousand. Physics tells us that it takes a structure – either neurological or electronic – of at least 100ms in length to achieve the pitch acuity of a human, and this structure is attuned to the onset – not the decay – of sounds. With some confidence we can predict that if at the onset of a signal the direct sound is stronger (in a neurological sense) than the sum of all reflections that occur in the first 100ms, we will be able to decode the information – pitch, timbre, and location – that we need to separate this signal from reflections. And with just a little more direct sound we can separate musical lines from each other.
So we know there is a time window, active at the onsets of sounds, that separates a sound field into independent neural streams by their fundamental pitches. We don’t need to know the details of this process to understand the effects of acoustics. A simple diagram will suffice. Our recordings of balloon pops in Rockport clearly show the arrival of the direct sound followed by multiple reflections. It is known that the ear perceives loudness – sound strength – as roughly proportional to the logarithm of sound pressure, which is why sound engineers usually graph sound pressure in decibels – defined as times the log of the sound pressure squared.
Using the balloon data from row G we can graph the direct sound pressure at the onset of a note separately from the build-up of reflections and reverberation. We assume the note is at a constant level, so the log of the direct sound pressure at the onset will also be constant. But the level of the reflections builds up as each additional reflection strikes the listener. If we adjust the balloon data so the level of the direct sound is the same for all three balloons, we get the picture shown in figure one. The balloon data has been band-limited to contain only frequencies between 700Hz and 4000Hz, as these frequencies contain most of the information in both speech and music.
The blue line in Figure 1 shows the constant level of the direct sound from our note. The red line shows the buildup of reflections when the glass window is uncovered. The zero of the vertical axis has been adjusted to correspond to the final total sound pressure when the window is uncovered and the note is held. We can see that when the glass is uncovered the direct sound is 9dB less than the final total sound pressure. The room has increased the loudness of a held note by about 9dB! This is a lot – probably more than is needed in such a small space.
The green line shows the build-up of reflections when the wooden shutters are in front of the window. These shutters absorb or redirect many of the strong reflections from the glass, so the buildup is not as fast or as strong as when the window is uncovered. The cyan line shows the buildup with a heavy fabric curtain covering the window. The fabric attenuates even more of the early reflections, leaving the direct sound unobstructed inside the 100ms window.
Conventional measurements of the late reverberation time (RT) – which measures how sound decays in a space, and not how the space affects the onsets of sounds – show that neither the shutters nor the curtain affects the decay of sound very much.
Late RT at 1000Hz with the window open = 1.1s
Late RT at 1000Hz with the wooden shutters = 1.0s
Late RT at 1000Hz with the fabric curtain = 0.9s
But as you can see in Figure 1, the window coverings affect the onset of the sound dramatically. A simple signal-to-noise argument suggests that the direct sound is separately audible if the area under the blue line in the 100ms window (the total number of nerve firings from the direct sound) is greater than the area under the other colored lines in the same window. (We assume the nerve firings stop at a sound pressure level of 20dB below the maximum total pressure when the note is held – which is neurologically plausible.)
In making this suggestion we are equating the rate of nerve firings to the log of the sound pressure – and then finding the difference in the total number for the direct sound and for the reflections. If there are more firings from the direct sound we will be able to extract information from it. If there are more from the reflections the direct sound will be obscured.
We can write some mathematics to calculate these areas and make a measure for clarity. For lack of a better name, I call this measure LOC, because the data I used to verify it came from localization experiments. If we calculate LOC for the three curves above, we get:
LOC with the window open = -2.0dB
LOC with the wooden shutters = +0.1dB
LOC with the fabric curtain = +3.8dB
Balloons are not the ideal sound source for measurement. They take three to five milliseconds to burst, spraying sound energy in random directions as the tear streams across the balloon surface. They are not very repeatable, and ideally one would average several. But they are quick, friendly to people in the hall, and the data is quite useful. Figures 2 and 3 show two balloon bursts as they were recorded by the Soundfield microphone, one with the glass window uncovered, and one with the wooden shutters in place. Allowing for the random differences due to the balloon bursts, one can see where the major reflections are coming from.
Figure 2: Soundfield microphone output with the glass uncovered. Red is the front/back direction, blue is left/right, and black is up/down. Note the two strong reflections off the window and side walls at 27 and 35 milliseconds.
Figure 3: Soundfield microphone output with the wooden shutters in place. Note the two prominent reflections in Figure 2 are re-directed or absorbed. The reverberant energy builds up more slowly.
Although the window coverings do not affect the rate of sound decay, we see that they do affect the clarity of onsets. As can be seen in Figure 1, they also affect the final loudness of held notes. The final loudness of a held note with the fabric curtain in place will be six or seven decibels lower than with the curtain open. The values of LOC predict that an omni-directional instrument playing in front of the glass window will not be clearly localized in row G in the unoccupied hall, and it would be clearly localized with the fabric curtain in place. An interesting result – but the real question is, what about a directional instrument in the occupied hall?
As Hawley was warming up on stage, I listened to various seats around the hall to get a sense of where to put the microphones. I was surprised at the differences in sound in the various seats. Yukiko Uneo-Egozy was doing the same thing – and there was considerable agreement between us about the differences we were hearing. In row F on the floor the low registers of the piano dominated the sound, and the left hand notes were blurry. In row G the sound was better balanced and clearer. The front row center seat in the balcony had a bright, warm sound, but again a blurry left hand. The fourth row in the balcony seemed both balanced and clear. I was able to verify these observations later from the recordings I had made. Why should the sound close to the piano be less clear than further away? And what is special about the first row of the balcony?
In row F a listener is below the piano, looking up at the underside. The sound of the treble strings bouncing off the open top goes over the heads of listeners in row F, but reaches listeners in row G and in the balcony. So in row F the low frequencies dominate. In addition, there is a strong reflection to row F from the glass window behind the piano. When this reflection (which is from the underside of the piano) combines with the direct sound, some of the clarity is lost.
Listeners in the first row of the balcony are exposed not only to the direct sound from the piano – which is well balanced – but also to reflected sound from below, above, left and right. That’s a lot of reflected sound, particularly when the hall is empty. Since the reflections come from all around, there is a spatial quality to the sound that is quite attractive. But when the hall is empty there is just too much reflected energy for a clear sonic image. Just four rows back in the balcony a listener is shielded from many of these reflections, and the sonic clarity is restored. I was not aware of it at the time, but the recordings made both in rehearsal and in the concert show that seats in balcony row four have distinctly less low frequencies than the seats in the front row of the balcony. Not really a problem. The clarity of the left hand is thereby increased. Why the low frequencies were reduced I do not know. It may be related to the “seat back effect” which is absorption primarily of low frequencies when sound travels over rows of seats. Our ears adapt to frequency balance. In Rockport the differences I noticed at low frequencies are small enough that they are unlikely to be problematic after a few minutes of listening.
The problems I heard during rehearsal disappeared when the hall was full. The added absorption – and there was a lot of it – reduced the reflected energy in all seats, increasing the clarity and reducing the differences between them. The major difference that remained was loudness. Naturally the sound was loudest in row F, and softest in the balcony row 4.
What about the effect of the unobstructed glass window? It may be a problem for some concerts, but it was not a problem for this one. The reason is simple: For most seats, the major reflections from the window were blocked by the piano body and the open cover. The clarinetist stood right in front of the open piano cover, which shielded his sound from striking the window. His position dramatically increased the directivity of his instrument, limiting the degree to which it excited the hall, and maximizing the strength of the direct sound. The result was an attention-grabbing sense of presence. Just what Brahms would want.
I deliberately popped the balloons in the open and not in front of the piano. The data reveals that the woven wooden curtain (and/or the fabric curtain) really does make a significant increase in clarity when there is an ensemble – such as a woodwind quintet – that is more or less omni-directional. A friend, who heard such a concert from the balcony with the window uncovered, said the clarity was less than optimal. The open window, however, is beloved by the audience, and significantly adds to the pleasure of the moment (especially during the golden hour.) Sound is one of the least important factors in the enjoyment of a concert, as a clear visual image can substitute for a clear sonic image. It is also nearly impossible while watching a concert live to concentrate on small details of what you are actually hearing. But for both the Parthenia Consort in row G and the Brahms concert, in at least four seats the hall worked well both sonically and visually.
David Griesinger is a Harvard-trained physicist who is eminent in the field of sound and music. His website is here.