Page 28 - Fall 2005

geometric spreading. It is modeled in terms of the sound power level of the source and the sound pressure level at the receiver. When the distance between the source and the receiv- er doubles, the sound level is reduced by 6 dB.
The noise in our simple example is the total contribution of all the talkers in the room. In this case we assume that this is the reverberant field level, or all the sound that has not come directly from a talker to the receiver. The reverberant level is the sound that has encountered the surfaces of the room one or more times, and it tends to be constant. Clearly, some of the reverberant field sound produced by our subject talker might fall into the signal category and some of the direct field sounds from other patrons might be considered noise. Although we acousticians have a good time arguing about definitions, for purposes of this analysis we will ignore these contributions. The reverberant field level shown in Eq. 2 is only dependent upon the total sound power Lw of all the talkers in the room and the room constant, which is the total amount of absorption due to all the surfaces of the room.
The cocktail party effect
How does a room get to be noisy? If there is a band present or other music is being played, this sound is treated as noise from the standpoint of understanding speech. Since there is not much we can do about these sources except turn them down, what we want to focus on here is the sound generated by the conversations between other patrons. There is a phenomenon called the cocktail party effect, which is an interesting and amusing exercise in the buildup of a sound field in a room. Let us assume that we are giving a party in a relatively reverberant room and invite a number of people to attend. The room has a carpeted floor, hard walls and ceiling, and some furniture, which contribute 93 metric (1000 sq ft) sabins of absorption (the A in Eq. 2). Before the guests arrive, the two hosts are hav- ing a conversation in the living room. They are polite so only one speaks at a time, each generating a sound power level of 70 dB. For the purposes of this calculation we assume that the direct sound transmitted between the talker (with Q = 2) and the listener, is the signal, and the reverberant sound reflected from the surfaces of the room is the noise. Clearly, some of the reflected sound contributes to intelligibility but we are going to ignore that for this simple analysis.
Using Fig. 1, for barely adequate
(60%) intelligibility, we need a signal-
to-noise ratio of at least –6 dB to
understand sentences.
The reverberant field level in
our living room is
This means that speech can be understood at a direct field level of 50.3 dB. Assuming the background noise due to other sources is low, two people can converse comfortably at a sepa- ration distance of 3.9 m (13 ft).
Our first guests arrive and two groups begin talking, only now two people, one from each group, are talking simul- taneously. The reverberant level increases by 3 dB (10 log N), but the direct field remains the same, so the minimum con- versation distance drops to 2.7 m (9 ft). When two more cou- ples arrive and pair off, the comprehension distance drops to 1.9 m (6 ft). When four more arrive the distance drops to 1.3 m (4 ft), and so forth.
In practice what happens is that people may choose not just to move closer, but also to talk louder. This raises the over- all background noise and forces everyone to elevate their voic- es so at the end of the evening they all go home with sore throats—a corollary of the cocktail party effect. The point of this example is that more absorption in the room yields a high- er signal-to-noise ratio and more people can talk comfortably before the increasing-volume spiral begins to kick in.
Restaurants
Restaurant design includes a similar problem in speech intelligibility since we want patrons to be able to talk com- fortably across a table, but we do not want their conversations understood by someone at a neighboring table. Consequently we need sufficient absorption so that we do not have to raise our voices at a cross-table distance of 1 to 2 m (3 to 6 ft), but we want masking at a table-to-table distance of, say, 3 m (10 ft) and beyond.
Let us imagine a restaurant that has a hard ceiling and walls and some absorption in the furniture for a total of, say, 20 metric sabins. A normal conversational level (Lw = 70 dB) will produce a direct field of 60 dB at 1.2 m (4 ft). With 20 metric sabins, our self-generated reverberant-field noise is 63 dB, our signal-to-noise ratio is - 3 dB, and we achieve 75 percent intelligibility. If there are 20 tables in the room, with one person talking at each table, the reverberant noise level rises by 10 log 20 to 76 dB, a very uncomfortable (4 bells) level, and we can no longer hold an intelligible con- versation. This simple calculation tells us something use- ful—in hard-surfaced restaurants it is very difficult to have a normal conversation across a table. People who enjoy con- versing with their dinner companions do not return to these establishments and the restaurant owners ultimately suffer. Yet for some unfathomable reason countless restaurants are designed in this way.
We address the problem by adding absorption (such as one-inch-thick fiberglass panels wrapped in cloth) to the walls and ceiling. Carpeting or other thin materials do very little. Now assume that we cover the ceiling with an absorbent materi- al. If it has an absorption coefficient of 0.9, this adds 170 metric sabins to the 13.7 x 13.7 m (45 x 45 ft) room. The 20 table rever-
“Form follows fashion rather than function in the world of architecture.”
26 Acoustics Today, October 2005