Classroom noise meter

Pick the activity, press Start the microphone and allow access when the browser asks. The light then follows the room: green while the level stays under the quiet limit, amber in a band just below it, red once the limit is reached. It turns red only when the room stays loud, because the default response ignores a single bang.

Each activity sets a starting limit, from 50 dB(A) for silent work or a test to 80 dB(A) for practical work or play, with 68 dB(A) as the default. Each sits at or a few decibels above the level that Shield, Greenland and Dockrell's 2010 review reports for that activity in primary classrooms. A class working as planned therefore stays green. The review gives no figure for practical work, so that limit is the site's own.

The amber band begins 5 dB below the limit and can be set anywhere from 1 dB to 20 dB. An own limit can be anything from 30 dB(A) to 110 dB(A). Set the limit from the room works the other way round. Pressed while the class works at a level the teacher accepts, it puts the limit 3 dB above the steady level of that moment.

A printed noise-o-meter is a paper dial whose pointer someone moves by hand; this traffic light moves by itself. Full screen fills a projector screen or an interactive whiteboard with the light on a dark stage, beside the text in landscape and above it in portrait. Where a browser cannot go full screen, as on an iPhone, the stage covers the page instead. The screen is kept awake while measuring wherever the browser supports it.

While the meter runs, Space pauses and resumes it and F opens or closes the large display. The up and down arrows nudge the limit one decibel at a time. A countdown of 5 to 60 minutes counts the time measured and pauses the meter at zero with the words "Time is up". Counting days instead of minutes, such as the days left until half term, is what the days until countdown does.

The warning sound is off by default. Switched on, it plays a short two-note chime each new time the light turns red, and phones that allow it vibrate as well. The meter ignores its own chime, so the alert never counts as noise, and the same chime sounds when a countdown ends.

Any classroom noise monitor that runs in a browser starts from the same raw material: blocks of samples from the microphone. This one asks the Web Audio API's analyser for the spectrum of each block of 4,096 samples, a fraction of a second of sound. The W3C specification has the analyser cut each block with a Blackman window before the spectrum is taken.

Each frequency in the spectrum is then weighted. A-weighting, defined in the IEC 61672-1 standard for sound level meters, follows the ear's weaker response to deep and very high sounds. It is -19.1 dB at 100 Hz, 0 dB at 1 kHz and +1.2 dB at 2 kHz. A level weighted this way is written dB(A). The Z setting leaves every frequency flat, for comparison with an unweighted reading.

The weighted spectrum is summed back into one level in decibels relative to digital full scale (dBFS), where 0 dBFS is the strongest signal the microphone can pass on. An offset turns that into a sound level. By default full scale is taken to stand for 120 dB, a typical figure for the microphones in phones and laptops.

How quickly the reading moves is set by its time weighting. Fast and Slow are the F and S responses of IEC 61672-1, with time constants of 0.125 s and 1 s. After a sound stops, Fast falls about 34.7 dB a second and Slow about 4.3. Steady, the default, is the site's own: the median of the fast level over the last 5 s, the level exceeded half the time in that window (L50).

Until it is calibrated, every figure is an estimate, and the display says so. A web page cannot learn its microphone's sensitivity, and datasheets show how much that varies. TDK InvenSense specifies its ICS-43434 at -26 dBFS for a sound of 94 dB, which puts full scale at 120 dB. Infineon's IM69D130, at -36 dBFS, puts it at 130 dB. The default follows the first; a phone's case and software shift it further.

Phones have been tested against reference meters. NIOSH researchers Kardous and Shaw (2014) looked at more than 130 iOS apps and tested 10, along with 4 of 62 Android apps. The closest came within 0.52 dB of the reference on the A scale, and a few more within 2 dB. Their 2016 follow-up, with calibrated external microphones, found apps within 1 dB. No comparable study of laptop or browser microphones was found.

Calibration replaces the default with a measured calibration offset. A calibrated sound level meter, set to A-weighting and slow, stands beside the device while a steady sound plays, such as rain noise from a speaker. Typing the meter's reading sets the offset to that reading minus the microphone's slow level in dBFS. An offset from a datasheet can be typed instead, from 60 dB to 160 dB, and a reset returns to 120 dB.

IEC 61672-1 sorts sound level meters into Class 1 and Class 2, with wider acceptance limits for Class 2. A browser meter carries no class, calibrated or not. A calibration also holds only for the device, browser and microphone it was made with, which is why it stays in the browser where it was saved.

Browsers can process a microphone for calls, which would bend a level reading. Start asks the browser to switch off echo cancellation, noise suppression and automatic gain control. The W3C's Media Capture and Streams specification notes that turning each off keeps the audio unaltered. Not every browser offers all three switches. When the browser does not confirm that all three are off, a note says the level may be less reliable.

No UK rule sets a noise level for a lesson in progress. Building Bulletin 93 (BB93), the Department for Education's acoustic standard for schools in England, limits the indoor ambient noise level of an unoccupied classroom. The upper limit is 35 dB LAeq over 30 minutes in a new building and 40 dB after refurbishment. That figure covers traffic, heating and ventilation with nobody teaching, so it leaves out the noise of the class itself.

Lessons measured in use are far louder. Shield and Dockrell (2004) surveyed 140 classrooms in 16 London primary schools: occupied, they averaged 72.1 dB LAeq, against 47 dB empty. Across 274 secondary lessons in 80 classrooms, Shield and colleagues (2018) measured 45 to 76 dB LAeq, with an average of 64.2 dB. A limit for a lesson is a teacher's choice, set against levels like these.

BB93 also caps the background in other rooms, again for a new building with nobody in it.

• Open plan teaching areas: 40 dB LAeq.
• Rooms for pupils with special hearing or communication needs: 30 dB LAeq.
• Dining rooms: 45 dB LAeq.

Reverberation time, how long sound lingers after it stops, is BB93's other measure. The mid-frequency figure may be at most 0.6 s in a new primary classroom and 0.8 s in a new secondary one. The World Health Organization (WHO) asks for no more than 35 dB LAeq of background noise during teaching and about 0.6 s of reverberation, for the sake of speech intelligibility.

In a noisy room people speak louder without deciding to, a reflex known as the Lombard effect. Bottalico, Passione, Graetzer and Hunter (2017) measured its strength: speakers added about 0.24 dB for each decibel of noise below roughly 43.3 dB(A), and 0.65 dB per decibel above it. A class that grows louder therefore pulls the teacher's voice up with it.

Speech must stand clear of the noise to be understood. The WHO's 1999 guidelines ask for a signal-to-noise ratio of at least 15 dB for complicated messages at school. Shield, Greenland and Dockrell's review adds that the youngest classes need a wider margin. By the WHO's figures, casual speech at one metre is about 45 to 50 dB(A), and shouting about 30 dB more.

For hearing, the US National Institute on Deafness and Other Communication Disorders (NIDCD) puts normal conversation at 60 to 70 dB(A). It says sounds at or below 70 dB(A) are unlikely to cause hearing loss, while sounds of 85 dB(A) and above can.

A voice also tires over long stretches of talk. How long a prepared explanation or a passage read aloud will run can be worked out from its word count with the speech time calculator.

The Control of Noise at Work Regulations 2005 protect employees, so in a school they cover teachers and other staff, not pupils. They measure the daily personal noise exposure (LEP,d): the level averaged by energy over a nominal working day of 8 hours. Northern Ireland has its own regulations of 2006 with the same values.

Regulation 4 sets the lower exposure action value at 80 dB(A), the upper exposure action value at 85 dB(A) and the exposure limit value at 87 dB(A). The action values are compared without hearing protection, and only the limit allows for the protection worn. Each also has a peak in dB(C), which this meter does not read, since it offers only the A and Z weightings.

The panel under the meter adds up to 4 parts of a day, each counted by its sound energy and by its share of those 8 hours. Its starting example, 5 hours in a lively classroom at 72 dB(A), gives 70.0 dB(A), below the lower value. A button copies the session's average into the first part.

HSE's guidance on the Regulations (L108) works with a 3 dB rule: half the time at 3 dB more gives the same exposure. So 83 dB(A) for 4 hours and 86 dB(A) for 2 hours both reach the lower value, like 80 dB(A) all day. Its rough tests put noise at about 80 dB(A) when it is intrusive but normal conversation is possible, and about 85 dB(A) when people shout to talk two metres apart.

Music teaching comes closest to the values. HSE's Sound advice (HSG260) gives singing lessons at 85 dB and piano lessons at 82 dB. At those levels the calculator finds 2.53 hours and 5.05 hours before a teacher reaches the lower value, close to HSE's own table. The same guide notes that damaging levels have been measured at the conductor's position in school bands.

Lessons and ordinary teaching add up. A day with 2 hours of singing at 85 dB and 3 hours in a classroom at 72 dB(A) comes to 79.3 dB(A). A third hour of singing, with two in the classroom, gives 80.9 dB(A), over the lower value. A device on a desk is not at the teacher's ear, so these are estimates. The reference cases run HSE's published examples through the same calculator.

Below the light, the session figures cover the time measured since the microphone started or the figures were last cleared. Average level (Leq) is the equivalent continuous sound level, written LAeq when A-weighted: the steady level that would carry the same sound energy over the same time.

Decibels are logarithmic, so the average leans towards loud moments. Each extra 3 dB doubles the sound energy, and an extra 10 dB multiplies it by ten. A few loud minutes can therefore lift the Leq well above the level heard most of the time.

Highest level (fast) is the loudest Fast reading of the session. The shares of green, amber and red show how the time divided, and the red share is the time over the limit. Times over the limit counts each new red: the light must spend at least 2 s out of red first, so a level hovering at the limit counts once. Longest stretch below the limit is the longest run without red, amber included.

A graph traces the last 2 minutes against the limit. Copy result puts a short text summary on the clipboard, for a lesson log or a report, and New session clears the figures for the next class.

Nothing the microphone hears is recorded, kept or sent. The sound feeds an analyser inside the page, which turns it into levels on this device, and only those levels are used. The browser shows its recording indicator while the microphone is on, and Turn the microphone off releases it.

The microphone permission belongs to the browser: it asks the first time, and the site settings beside the address can withdraw it. Under the W3C's Media Capture and Streams specification only secure (https) pages may ask at all, and on this site only the noise meter's pages are allowed to.

Just one thing stays behind: a saved calibration, with its offset, date and method, in this browser's local storage until it is reset or the site's data is cleared. A link to the page carries the settings and the exposure parts, never any sound.