Measuring Reverberation Time: An Essential Character Of Room Acoustics

Signal-To-Noise Ratio

The ISO standard is defined for 30 dB of room decay, and extrapolated to 60 dB. The reason is that actual impulsive testing tends to produce RIRs with poor signal-to-noise ratio, so one can rarely get the required decay before running into the noise floor.

The presence of noise causes a reduction in the slope of the late part of the Schroeder curve (Figure 5). It can be impractical to expect a quiet environment when testing real spaces (Sabine did his testing in the middle of the night), where Murphy’s Law assures that a vacuum cleaner is likely running somewhere in the vicinity.

In fact, it can be difficult to get 35 dB of decay. The Standard provides a T20 (-5…-25) and T10 (-5…-15) that can be used in lieu of T30 for noisy data, both extrapolated to 60 dB of decay.

Figure 5: Noise can corrupt the determination of T30. The 8 kHz Log2 RIR below shows significant noise. Manual cursor placements are necessary to determine the RT.

Measurement systems used for testing sound systems can easily yield 60 dB or more of SNR, so it is not necessary to base the RT on only 30 dB of decay. The slope-change at the end of the Schroeder curve is most often due to noise being integrated along with reverberation.

Most measurement programs provide automatic or manual methods for excluding this part of the curve from the calculation. But, what if it’s not noise?

Multi-Sloped Rooms

Some rooms can actually have a triple-sloped Schroeder curve. This can happen in “coupled spaces” where two different sized rooms are joined.

The first 10 dB (the basis for EDT) is steep due to a strong direct field and/or strong early reflections. A lower-slope straight-line section is next (for one space) and a lower-yet slope segment is next for the more reverberant of the two coupled spaces.

These slope changes can cause problems for automatic T30 algorithms that are programmed according to the Standard. Again, manual cursors can help handle these special cases.

How does one know if a late slope change is due to noise or a multi-sloped room? A good noise compensation algorithm can often make the distinction, but no standard exists so this is left up to the ingenuity of the programmer.

Interpretation Of The Standard

Another source of differences between measurement platforms lies in how the Standard is interpreted. The Standard is based on a “best fit linear regression line” to the Schroeder curve and should be “free” at both ends.

Figure 6 – Curve fit to the EDT having two degrees of freedom.

A line (“fixed” at the left end) connecting the relevant levels (0 to -10 dB for EDT, -5 to -35 dB for T30) is not the Standard but is used by some measurement platforms. Figure 6 shows an EDT determined from a straight line that is free on both ends.

Preferred Measurement Systems

This should all serve as a reminder that the measurement and interpretation of acoustical data requires some judgment. Here are some things to remember:

– Even though Standards exist, they may have been written based on assumptions that are not necessarily realized or dominant when a sound system is involved.

– “One number” specifications are rarely useful. Practitioners should gather a complete, low-noise RIR at any seat that is worth investigating.

—Data should be platform-independent to allow processing in multiple measurement programs. The WAV file has emerged as a popular choice. If you are in doubt about a particular metric, opening the same RIR in a different program can yield some insights.

– Metrics such as T30 are one piece of a large, complex puzzle. Additional metrics such as C50 and EDT provide other pieces.

One of the best ways to evaluate an RIR is by convolution. GratisVolver is a powerful near real-time convolver for listening to the RIR as well as deconvolving wet sweeps. Thus, the “final word” may come from the most popular and potentially powerful analyzer of all – the human ear/brain system.