Method

The Loudspeaker Check predicts the signal-to-noise ratio a BMR measurement will reach and compares it with what the BMR analysis requires. The constants come from the literature, from the BMR analysis, or from BMR's stored measurements.

That a reduced chain — a loudspeaker of this kind and a phone — can measure a room at all was tested against professional reference equipment in parallel measurements of small to medium-sized meeting rooms, evaluated in the master thesis of B. Pohler, 2025: given enough signal-to-noise ratio and no strong non-linearities it gives usable reverberation values in the mid frequencies, and clear limitations remain at low frequencies.

Symbols

Symbol	Quantity	Unit
$L_{1\mathrm{m}}$	Datasheet maximum level, referred to 1 m	dB SPL
$L_{\mathrm{sw}}$	Broadband sweep level at 1 m, full volume	dB SPL
$L_{\mathrm{mic}}$	Sweep level at the microphone	dB SPL
$L_A$	Background noise, A-weighted	dB(A)
$L_{N,b}$	Background noise in octave band b	dB SPL
$A,V,d$	Floor area, volume, source–microphone distance	m², m³, m
$T$	Reverberation time	s
$Q$	Directivity factor, 2	—
${\mathrm{SNR}}_b$	Signal-to-noise ratio in band b	dB
$M,\beta ,\sigma$	Margin to the threshold; borderline zone; uncertainty of the datasheet level	dB

What the measurement requires

The BMR analysis needs a minimum signal-to-noise ratio in each octave band. The reverberation time is built from the 500 Hz and 1 kHz bands, so the lower of the two decides.

Tier	SNR	Below it
Minimum	25 dB	no reverberation time; the app asks for a re-record
Works	28 dB	no margin: a few dB less and the measurement fails
No noise warning	35 dB	"close to noise floor" warning
T30 available	38 dB	no T30

25, 35 and 38 dB are where the BMR analysis acts. 28 dB is this tool's own line: the minimum plus 3 dB, so that "works" does not mean "only just". Between 28 and 35 dB the analysis may warn that the signal is close to the noise floor. ISO 3382-1 and -2 ask for a decay that starts at least 35 dB above the noise for T20 and 45 dB for T30; that is a level at the start of the decay, which is not the same quantity as the average signal-to-noise ratio used here. In BMR's stored measurements a reverberation time measured between 28 and 35 dB agrees with the same room measured well above it, inside the scatter between repeat recordings, though that scatter widens towards the top of the range; below 28 dB the few recordings there disagree, sometimes by a wide margin. The 28 dB line is a practical choice, not a standard.

Full result: the bands the MOS score needs

The headline room size is the largest room in which the measurement gives the full result: 500 Hz and 1 kHz at 28 dB and 125 Hz and 250 Hz at 25 dB. The MOS score's Bass Ratio, $(T_{125}+T_{250})/(T_{500}+T_{1000})$, needs a reverberation time in all four bands, and a band below 25 dB returns none. Just above that line it can still be missing. The tool keeps its caution in the loudspeaker level and the far-corner microphone rather than adding a third margin here, so the Bass Ratio is not certain at the headline size. Checked back against BMR measurements, most recordings in rooms the tool passes do return the full recording; where one does not, it is usually the Bass Ratio that is missing rather than the reverberation time.

Loudspeakers also arrive weaker at 125 Hz and 250 Hz than a flat model assumes. BMR's calibrated recordings set that shortfall, which is applied to every loudspeaker, and its spread widens the uncertainty of those two bands — most of that width being the room-mode spread at 125 Hz, a property of the room rather than of the loudspeaker.

The color of an answer therefore follows the measured failure rate, not that width: a predicted 28 dB or more in the Bass Ratio bands is not marked down, below it the answer turns amber and says the band may drop out. Marking down on the width instead would put most loudspeakers in amber. It still sets the size range and the tiers, and a missing Bass Ratio is never red: the reverberation time still comes back.

If 125 Hz or 250 Hz is outside the loudspeaker's range, for example a speakerphone in its voice mode, the MOS score has no Bass Ratio and the answer says so. The room size stays the one its level allows with that band played: switching a band off does not make a room bigger.

The range given with the room size shows how far the size moves when every uncertain part (the model's error against stored measurements, the unknown sweep level, the spread of the low bands) is moved to its cautious or its optimistic end. It is not a probability. In large rooms a few decibels change the size several-fold, so the range is wide.

The check covers rooms up to 500 m³: DIN 18041 states its A4 target, which lists the video-conference room as an example, for rooms up to that volume and calls A4 unsuitable above it. As a floor area that is 500 m³ divided by the ceiling height. The reverberation time is still estimated with the A3 target, as BMR's analysis does.

From the datasheet to the microphone

$$L_{1\mathrm{m}}=L(d_{\mathrm{ref}})+20{\log }_{10}{d}_{\mathrm{ref}}$$

(2)

$$L_{\mathrm{sw}}=L_{1\mathrm{m}}+k,k=-12\mathrm{dB}$$

(3)

Some datasheets give a wattage and no maximum level at all. The level is then estimated from it:

$$L_{1\mathrm{m}}=s_W+10{\log }_{10}W,s_W=74.2\mathrm{dB}$$

(2a)

$s_W$ is not a driver sensitivity. A driver's 1 W / 1 m figure is measured at small signal and ignores the compression, limiting and excursion limits that decide the maximum. Where BMR has recordings of such a loudspeaker, an assumed 80 dB per watt sized it at least 8 dB high. It is read off finished products instead, from the datasheets that publish a maximum level, the distance it was measured at and a wattage. The median of those is used and their spread, 5.8 dB, is taken off the level, so a loudspeaker sized this way is an estimate and the answer says which of the two routes it took.

Checked against BMR recordings of a portable that is not in that set, equation (2a) lands inside the range the loudspeakers with a published maximum level sit in. That is evidence it is not 8 dB high, as an assumed driver sensitivity was, rather than evidence that it is right: the wattage term is anchored rather than tested, and a published maximum level is always the better input.

$k$ is the level a sine sweep reaches at full volume relative to the maximum-SPL rating, and the least certain number here. Manufacturer ratings are mostly pink noise at full drive, reported as the highest peak (Weinzierl, ch. 17); a sweep has a 3 dB crest factor, so crest factor alone would put $k$ near −3 dB. In BMR's recordings of loudspeakers with a published level, played at an unknown volume, the offset per device lay between −15.4 and +0.3 dB. The −12 dB used is a cautious choice from that range, and a full-volume measurement per device will replace it.

$$d=\sqrt{A/2}$$

(4)

Source centred, microphone in the farthest corner of a square room.

$$T=a{\log }_{10}V+b\text{(DIN 18041, A3: }a=0.32,b=-0.17)$$

(5)

Used when the reverberation time has not been measured. In BMR rooms the measured value sits a little under this target on average, so it is a fair estimate of these rooms.

$$L_{\mathrm{mic}}=L_{\mathrm{sw}}-10{\log }_{10}\frac{Q}{4\pi }+10{\log }_{10}\left(\frac{Q}{4\pi d^2}+\frac{25T}{V}\right)$$

(6)

Direct sound plus the reverberant field. Both reach the microphone, so both count. The reverberant term $25T/V$ is the classical $4/A$ with Sabine's $A=0.161V/T$, since $4/0.161=24.8$. Where only the microphone moved, BMR's measured level fell far less with distance than the direct sound alone predicts (slope near 0 against −1). Equation (6) predicts −0.3 to −0.5 there, so its distance dependence is not confirmed either.

$Q$ is 2, and that is a deliberate match rather than a measurement: the BMR analysis uses the same value, where it stands for talker directivity. A loudspeaker is not a talker, but what this page predicts is what that analysis will return, so the two agree by construction and a disagreement between them cannot come from $Q$. What it costs is in the next section.

The SNR the measurement will reach

$$L_{N,b}=L_A+\Delta N_b$$

(7)

Band	125 Hz	250 Hz	500 Hz	1 kHz	2 kHz	4 kHz	8 kHz
ΔN (dB)	0.1	-1.7	-4.7	-7.2	-8.7	-11.2	-13.9

Median over BMR's calibrated measurements; room noise is highest in the low bands.

$${\mathrm{SNR}}_b=L_{\mathrm{mic}}-L_{N,b}+G,G=-3.9\mathrm{dB}$$

(8)

$G$ is a fixed property of the BMR measurement signal and its analysis, checked on simulated rooms to within ±1 dB from 500 Hz to 8 kHz.

$$M=\min ({\mathrm{SNR}}_{500},{\mathrm{SNR}}_{1000})-{\mathrm{SNR}}_{\mathrm{req}}$$

(9)

$$\text{works if }M\ge \beta ,\text{borderline if }|M|<\beta ,\text{does not work if }M\le -\beta ,\beta =6.5\mathrm{dB}$$

(10)

$\beta$ is how far the prediction can be trusted: 6.5 dB. It combines two parts that do not overlap. The first is the model's own error, the scatter left when the prediction is held against BMR's stored measurements with each session given its own loudspeaker level, so that the unknown volume setting is not counted here and again in $k$. Using each recording's own measured reverberation time instead of the DIN 18041 target makes it worse, not better, so the target is not the weak link. The second part is the unknown sweep offset $k$, spread over the −15.4 to +0.3 dB range above. It shrinks once $k$ is measured per device.

That first part is measured in small meeting rooms, not at the sizes where the headline matters most, and on portable loudspeakers and laptops rather than the conference systems this check is mostly used for. What is left over still tracks the background level, so the octave-band noise offsets of equation (7) carry a shape the model does not.

$\beta$ is also a floor rather than the whole error. One thing is deliberately outside it: the volume the loudspeaker is actually played at. The same loudspeaker measured on different days varies beyond everything above, and $k$ cannot contain that because it is built from one session per device. It is left out because the volume setting is the one part the person measuring controls.

$\sigma$ is subtracted from the loudspeaker's level when the datasheet leaves something open: 3 dB when it omits the measuring distance, 5.8 dB when the level was estimated from a wattage by equation (2a). Nothing is subtracted from the prediction, so the caution sits in one place and is visible in the working.

Where the sizes stop being checked at all: most BMR rooms are small, few are larger than 45 m², and the largest measured is 140 m². The check will state sizes above that — the volume cap allows 175 m² at a 2.8 m ceiling — and every answer that does says so in "what this rests on".

"No noise warning" means none in the 500 Hz and 1 kHz bands. Other bands can still warn, most often 125 Hz, and the analysis is stricter when the microphone is very close to the loudspeaker.

The largest floor area for a tier is the $A$ at which $M=0$. Beyond the reverberation radius the level falls only with $10{\log }_{10}(T/V)$, so a few decibels change that area several-fold: treat it as a guide and the verdict for the actual room as the answer.

Frequency coverage

A datasheet's frequency limit is where the output is down by a few dB (usually 3, sometimes 10), not where it stops. A band therefore counts when its centre $f_c$ lies inside the stated range, and is marked partial when the limit falls between $f_c/\sqrt{2}$ and $f_c\sqrt{2}$: measured, but weaker. BMR's measurements show both sides of that: room systems rated from 100 Hz reach the 125 Hz band but can land on the gate, and a speakerphone rated from 250 Hz falls well below it. A partial band on a driver smaller than 60 mm is flagged. The BMR app reports 125 Hz – 8 kHz; 63 Hz is not measured, because the sweep starts at 80 Hz.

What it does not include

• Directivity other than $Q$ = 2. The datasheet level is on-axis, so at a fixed on-axis level the power into the room falls as $Q$ rises: a loudspeaker of directivity $Q^{\prime }$ fills the reverberant field $10{\log }_{10}(Q^{\prime }/Q)$ dB less than equation (6) assumes, 3 dB at $Q^{\prime }=4$ and 5 dB at $Q^{\prime }=6$. A forward-firing soundbar is that case, and the check is optimistic for one. Modelling it needs the loudspeaker's directivity per octave band.
• A reverberant field that falls with distance from the source. Equation (6) treats $25T/V$ as the same everywhere in the room. BMR's recordings do not show that fall, so correcting for it would move the model away from the data.
• The loudspeaker's own frequency response, beyond the low-band shortfall above.
• Rooms that are not diffuse, where the model behind equation (6) does not hold. EN 12354-6 clause 5 notes that for its own absorption model the real reverberation time can then be up to twice the prediction.
• Rooms that are long and narrow. The tool assumes a square room.
• The volume actually set, limiters, and battery state. This is what $k$ stands in for.

Room presets

Class	People	Area	Height	Volume	Far corner
Huddle room	2–4 people	25 m²	2.9 m	73 m³	3.5 m
Meeting room	5–8 people	35 m²	2.8 m	98 m³	4.2 m
Conference room	9–20 people	60 m²	2.8 m	168 m³	5.5 m
Large meeting space	21–50 people	100 m²	3 m	300 m³	7.1 m

p75 of floor area per capacity class in BMR's room database; height is the class median.

Worked example

Teufel MYND in a meeting room, 35 m², 35 dB(A), 28 dB line.

What the measurement needs

1. SNR required (Works)
   $${\mathrm{SNR}}_{\mathrm{req}}=\text{28 dB}$$

What reaches the microphone

1. Datasheet level at 1 m
   $$L_{1\mathrm{m}}=95-20{\log }_{10}\frac{1}{1}=\text{95 dB}$$
2. Sweep level at full volume, conservative
   $$L_{\mathrm{sw}}=L_{1\mathrm{m}}-12=\text{83 dB}$$
3. Farthest microphone position
   $$d=\sqrt{A/2}=\sqrt{35/2}=\text{4.18 m}$$
4. Reverberation time, DIN 18041 A3
   $$T=0.32{\log }_{10}98-0.17=\text{0.47 s}$$
5. Sweep level at the microphone, direct + reverberant
   $$L_{\mathrm{mic}}=83-10{\log }_{10}\frac{2}{4\pi }+10{\log }_{10}\left(\frac{2}{4\pi 4.18^2}+\frac{25\cdot 0.47}{98}\right)=\text{82.1 dB SPL}$$

The SNR the measurement will reach

1. Background noise, 500 Hz band
   $$L_{N,500}=35-4.7=\text{30.3 dB}$$
2. SNR, 500 Hz band
   $${\mathrm{SNR}}_{500}=L_{\mathrm{mic}}-L_{N,500}-3.91=82.1-30.3-3.91=\text{47.9 dB}$$
3. Background noise, 1 kHz band
   $$L_{N,1000}=35-7.2=\text{27.8 dB}$$
4. SNR, 1 kHz band
   $${\mathrm{SNR}}_{1000}=L_{\mathrm{mic}}-L_{N,1000}-3.91=82.1-27.8-3.91=\text{50.4 dB}$$
5. SNR, 125 Hz band (Bass Ratio)
   $${\mathrm{SNR}}_{125}=L_{\mathrm{mic}}-L_{N,125}-3.91-3.6=82.1-35.1-3.91-3.6=\text{39.5 dB}$$
6. SNR, 250 Hz band (Bass Ratio)
   $${\mathrm{SNR}}_{250}=L_{\mathrm{mic}}-L_{N,250}-3.91-1.6=82.1-33.3-3.91-1.6=\text{43.3 dB}$$
7. Margin to the threshold
   $$\min ({\mathrm{SNR}}_{500},{\mathrm{SNR}}_{1000})-{\mathrm{SNR}}_{\mathrm{req}}=47.9-28=\text{+19.9 dB}$$
8. Borderline within
   $$\pm 6.5=\pm \text{6.5 dB}$$
9. Octave bands inside the speaker's range
   $${\log }_2\frac{\min (20000,11314)}{\max (52,88.4)}=7.00\text{of}7\text{octaves}(\text{7 of 7 bands})$$

Sources

1. BETTERMEETINGROOMS measurement analysis.
   The signal-to-noise ratio the analysis requires per octave band: 25 dB for a reverberation time at all, 35 dB for no noise warning, 38 dB for T30. The 28 dB line is this tool's own margin above 25 dB.
2. BETTERMEETINGROOMS measurement data.
   Stored room measurements used to set the octave-band noise offsets, to check the room-level model and the reverberation-time estimate, and to size the model's uncertainty.
3. Weinzierl (ed.), Handbuch der Audiotechnik, 2nd ed., ch. 11 (Ahnert & Weinzierl), Eq. (24), printed p. 226.
   Direct plus reverberant sound pressure in a room, for an omnidirectional source: the direct term falls as 1/(4πr²) and the reverberant term is the classical 4/A, which with Sabine's A = 0.161·V/T is 25·T/V. The directivity factor Q in the equation this tool uses is not in Eq. (24); it is the tool's own, set to 2 to match the BETTERMEETINGROOMS analysis.
4. DIN 18041:2016-03, 4.2.3, equations (1)–(6).
   Target reverberation time T_soll = a·log10(V) + b per usage type, used when the room has not been measured.
5. Weinzierl (ed.), Handbuch der Audiotechnik, 2nd ed., ch. 17 (Goertz & Makarski), printed pp. 430–431.
   Manufacturer maximum-SPL figures are mostly measured with pink noise at full drive and reported as the highest peak; a sine sweep has a crest factor of 3 dB. The tool assumes a sweep reaches 12 dB below the rating, which is conservative and will be replaced by full-volume measurements.
6. EN 12354-6:2003, (4.6) and (5).
   Validity of diffuse-field predictions: in rooms that are not diffuse the real reverberation time can be up to twice the prediction of that standard's model.
7. Bang & Olufsen, Beosound Explore product sheet, 02.2022 V03.
   Max loudness at 1 m 91 dB SPL; frequency range 56 – 22 700 Hz; 2 × 30 W Class D; 2 × 1.8" full-range drivers.
   https://assets.ctfassets.net/8cd2csgvqd3m/18ihLlRE5rxB5S2NIDpPCi/e0382bc943290a0ededb07bb530be8ee/Beosound_Explore_Product_Sheet_English_0001_Feb2021.pdf
8. Jabra Speak2 75 technical specifications, Rev C, 2024-09-17.
   Peak audio output 88 dB SPL at 0.5 m; bandwidth 80 Hz – 20 kHz in music mode, 150 Hz – 14 kHz in speak mode.
   https://www.jabra.com/_/media/Jabra_VXi_Product-Documentation/Jabra-Speak2-75/Technical-specifications/RevC/EN-Speak2-75-Tech-Spechs-180924.pdf
9. Teufel MYND, "Technische Daten", teufel.de product page.
   Maximaler Schalldruck 95 dB/1m, Frequenzbereich 52 – 20 000 Hz, 35 W RMS.
   https://teufel.de/mynd-107002004
10. Pohler, Room Acoustic Analysis and Material Estimation Using Reduced Measurement Setup, Master thesis (MEng), supervisors Prof. Dr. F. Melchior (Stuttgart Media University) and Univ. Prof. Dr. techn. A. Sontacchi (Institute of Electronic Music and Acoustics, University of Music and Performing Arts Graz), 2025.
    Parallel measurements of real meeting rooms with professional reference equipment and a smartphone and Bluetooth loudspeaker chain; finds the advantages and the limitations of a mobile setup.
11. Derived in this tool, from the datasheets listed on the method page.
    Level per watt at 1 m for a loudspeaker with no published maximum, the median of the datasheets here that publish a maximum level, its distance and a wattage (69.2, 73.2, 74.2, 79.6, 84.0 dB per watt). Also the octave band edges, f_c/√2 … f_c·√2.