Presets
Playback
Room and talker
Numerical checks
The science
What is and is not modelled. The title this page replaces was about "reality communication". What is modelled instead is a concrete, measurable thing: how well speech from one person can be understood by another in a room, depending on distance, reverberation and background noise. Nothing here models meaning, language, hearing loss or telepathy.
Sound in a box-shaped room
V = LWH, S = 2(LW + LH + WH), A = S·α RT (Eyring) = 0.161 V / (−S ln(1 − α)) RT (Sabine) = 0.161 V / A room constant R = A / (1 − α) critical distance rc = 0.141 √(Q R) (Q = 2) direct level Ld = L1m − 20 log r reverberant level Lr = L1m − 20 log rc reverberant/direct energy ratio ρ = r² / rc²Beyond the critical distance the reverberant field dominates, so moving further away no longer lowers the speech level but keeps smearing it. The speech level Ls is the energy sum of both parts and the signal-to-noise ratio is SNR = Ls − N against steady ambient noise N.
Single-band speech transmission index
Speech is a carrier modulated at roughly 0.6–12.5 Hz. A room and noise reduce the depth of those modulations by the modulation transfer function (Schroeder; Houtgast & Steeneken). For a direct sound plus an exponentially decaying reverberant tail (energy decay time RT) and noise:
m(F) = | 1 + ρ / (1 + j 2πF·RT / 13.8) | / (1 + ρ) · 1 / (1 + 10−SNR/10) SNRapp = 10 log[m / (1 − m)] clipped to ±15 dB, TI = (SNRapp + 15) / 30 STI ≈ mean of TI over the 14 modulation frequenciesThis page uses one broad band with equal weights, so it is an STI-style index for teaching, not the standardised seven-octave-band measurement. The words (bad < 0.30 < poor < 0.45 < fair < 0.60 < good < 0.75 < excellent) are the usual qualitative bands, quoted from memory.
Checks shown on the page
- The closed-form m(F) vs a numerical evaluation of the defining integral of the squared impulse response (40 000 time steps) at all 14 frequencies.
- Ray-tracing Monte Carlo (6 000 specular rays; each wall hit multiplies the energy by 1 − α): mean wall hits per second vs the mean-free-path law
c S / 4V. - The same Monte Carlo in a 6 m cube (a nearly diffuse case) at the current α: measured T20 reverberation time vs Eyring.
Honest limits
- Eyring and Sabine assume a diffuse sound field. In an empty rectangular room with specular walls the late decay is slower: the ray test for your room is shown as information (not a pass/fail), and its ratio to Eyring is often 1.3 or more.
- One absorption coefficient for all surfaces and frequencies; no furniture, no early-reflection weighting, no speech spectrum, no masking by other talkers, no level-dependent speech effort (people raise their voice in noise), no hearing loss.
- Voice level at 1 m (about 60 dB for normal speech) and the 0.161 constant (air at about 20 degrees C) are typical values, not constants; check them against an acoustics reference before quoting.
The 2D views
The plan shows the room from above: rings around the talker are coloured by the STI-style index at that distance, green lines are the direct path, orange lines the four first-order wall reflections (image sources), the violet dashed circle is the critical distance. Below it a level diagram plots direct, reverberant and total speech level and the noise against distance. New check: at the critical distance the direct and reverberant levels are equal and their sum is 3.0103 dB above the direct level (energy doubles).