Sound Transmission Loss

Transmission loss is the decibel statement of a plain fraction. A partition passes some proportion $\tau$ of the sound energy striking it — the transmission coefficient — and TL is $10\log_{10}(1/\tau)$. A wall at 30 dB passes a thousandth of the incident energy, at 40 dB a ten-thousandth, at 50 dB a hundred-thousandth. Each 10 dB is another factor of ten, which is why the numbers stay small while the performance changes enormously.

The measurement behind a published figure is specific and controlled. Under ASTM E90 or ISO 10140 a specimen is built into the opening between two isolated reverberation chambers, sound is generated in one, levels are measured in both, and the receiving room's absorption is accounted for so that the result belongs to the specimen and not to the rooms. Everything about that arrangement is better than a building: the specimen is sealed, it is not carrying structural loads, and there is no path around it. Field ratings — ASTC in the field versus STC in the lab — routinely come in several decibels lower, and the gap is almost never the wall's fault.

It is usually flanking. Sound reaches the next room along the continuous floor slab, through the ceiling plenum over the top of a partition that stops at the suspended ceiling, along a continuous window mullion, through back-to-back electrical boxes in the same stud cavity, or through the duct that serves both rooms. A partition can be perfect and the room still fail, because the flanking path does not care what the wall is made of. Any serious sound-isolation design deals with the paths around the partition first and the partition second.

Finally, TL is a curve, not a number. It rises with frequency, dips at the panel's coincidence frequency, and behaves badly near resonances, so the single-figure ratings — STC in North America, Rw in ISO countries — are contour-fitting exercises that compress that curve into one integer. Both are weighted toward speech frequencies, and both can rate a partition well while it performs poorly against bass from a subwoofer, a compressor or a bus. Where the source is low-frequency, ask for the octave-band data; the single number was designed for a different question.