How the tests work

Every AudioTestHQ test runs in your browser through the Web Audio API. Tones are generated live, not streamed. Measurements use your microphone through an AudioWorklet that reads raw samples. Nothing is uploaded. This page lists each signal and method so you can judge the results for yourself.

Safety

• Volume starts at 20% of the site's own level control.
• Every sound fades in and out over 30 ms, so nothing clicks.
• All stereo output passes through a limiter (threshold −3 dBFS, ratio 20:1) so peaks stay below full scale.

Signals

Measurements

• Microphone levels: peak and RMS over 100 ms blocks, in dBFS. Noise floor is the 20th percentile of block levels while you stay quiet; voice level is the 80th percentile while you speak; clipping counts samples at or above 99.9% of full scale. Browser echo cancellation, noise suppression and automatic gain are off unless you switch them on.
• Latency: clicks are scheduled at exact AudioContext times. Your microphone stream is captured with frame-accurate timestamps, and each click is found by cross-correlating the recording with the click shape in a 500 ms window. We report the median of the clicks found, and the spread between the 25th and 75th percentiles. A click only counts if its correlation peak stands clearly above the background.
• Sound level (decibel meter, classroom meter): 4,096-sample Hann-windowed FFT every 2,048 samples. The power spectrum is multiplied by the A, C or Z weighting from the analytic expressions in IEC 61672-1 (A and C normalised to 0 dB at 1 kHz) and summed for the weighted mean square. Fast (0.125 s) and Slow (1 s) exponential time weighting; Leq is the energy mean of all blocks; L10/L50/L90 come from a 0.1 dB histogram of the Fast level; peak is the largest sample, unweighted. Levels are dBFS plus an offset: 100 dB by default (uncalibrated), or the offset you set against a real meter, stored in your browser. In our tests with synthetic tones, 1 kHz read identically in A, C and Z; 100 Hz read 0.05 dB from the IEC A-weighting value; 31.5 Hz read about 1.5 dB high in A-weighting; a 10 s/30 s two-level sequence gave Leq within 0.01 dB of the calculated value.
• Spectrum analyzer: Hann-windowed FFT (2,048 to 32,768 points) of the newest samples, with optional exponential power averaging and peak hold. Line levels are in dB relative to a full-scale sine; the peak uses parabolic interpolation. 1/3-octave bands sum the power of FFT lines between 10^±1/20 of base-ten centre frequencies (1000·10^k/10), on the same scale as the decibel meter. In our tests, a 1 kHz tone at −6.02 dBFS was located within 0.4 Hz at all sizes and read about −5.9 dB at its peak; band and total levels matched calculated values to 0.01 dB; A and C weighting matched IEC 61672-1 at 100 Hz.
• Tone generator and sweep: Web Audio OscillatorNodes (band-limited sine, square, triangle, sawtooth), each with its own gain and stereo panner, faded in and out over 10 ms. Sweeps use exponential (logarithmic) or linear frequency ramps. WAV files are rendered from the same graph with an OfflineAudioContext, 16-bit stereo. In our tests, rendered tones and sweep end points matched the set frequencies when measured by FFT.
• DTMF: frequencies from ITU-T Q.23 (697/770/852/941 Hz and 1209/1336/1477/1633 Hz). The generator sums the two sines with 5 ms raised-cosine edges. The decoder uses the Goertzel algorithm on 25 ms windows every 10 ms, probing each tone at nominal and ±0.9%/±1.8%; a key needs one dominant tone per group, twist within 10 dB, and 60% of window energy, present for about 40 ms, with a 30 ms pause before a repeat (ITU-T Q.24 receiving values).
• Noise dose: OSHA 29 CFR 1910.95 Appendix A formulas (T = 8/2^((L−90)/5), D = 100·ΣC/T, TWA = 16.61·log10(D/100) + 90) and the NIOSH limit of 85 dBA with a 3 dBA exchange rate (T = 8/2^((L−85)/3)). Levels from 80 dBA up are counted. Checked against OSHA Table A-1 (50% = 85.0, 200% = 95.0, 10% = 73.4 dBA) and Table G-16 (95 dBA for 4 h = 100%).
• Pitch: YIN fundamental-frequency estimation on the latest 2,048 samples plus enough history for 40 Hz, about 23 times a second, with a cumulative-mean-normalised difference threshold of 0.12 and parabolic interpolation. Readings below 0.8 clarity or −50 dBFS are discarded. A note counts as held after 0.3 s at the same nearest note. In our test, synthetic tones with five harmonics and low background noise at A2, C4, E4 (+20 cents) and A5, fed to Chrome as a microphone, were each named correctly; median error under 1 cent, worst single reading 4 cents.
• Surround channel count: read from the browser's destination.maxChannelCount. Channels are then driven individually with discrete channel interpretation.

What a browser cannot measure

• True loudness in your room (dB SPL): needs a calibrated microphone. Our microphone readings are digital levels (dBFS).
• Your hearing: these are not hearing tests.
• Speakers behind a stereo-only connection: if your system reports 2 channels, we cannot reach individual surround speakers. Use the test files.

Sources

• W3C, Web Audio API specification (AudioContext, AudioWorklet, channel interpretation).
• Yale University Library, Music cataloging: vocal ranges (New Harvard Dictionary of Music ranges; Grove Music Online definitions for comparison), used for the vocal range test.
• Encyclopaedia Britannica, "countertenor" (definition and history), for the countertenor page.
• Lumen Learning, Music Appreciation, "Vocal Types and Ranges" (typical individual range of about one and a half to two octaves or more).
• de Cheveigné, A. and Kawahara, H. (2002). YIN, a fundamental frequency estimator for speech and music. Journal of the Acoustical Society of America, 111(4), 1917–1930.
• CDC National Center for Environmental Health, “What Noises Cause Hearing Loss?” (last reviewed 8 November 2022; withdrawn), Internet Archive copy of 2 June 2023, for sounds from 0 to 60 dB.
• US Federal Trade Commission, actions and warning letters on ultrasonic pest-control device claims; National Institute on Aging, “Hearing Loss: A Common Problem for Older Adults”; NIDCD, “Age-Related Hearing Loss” (last updated 17 March 2023).
• ITU-T Recommendation Q.23, Technical features of push-button telephone sets (DTMF frequencies and tolerance), and Q.24, Multifrequency push-button signal reception (receiving timing values).
• ISO 266:1997, Acoustics: Preferred frequencies (confirmed 2023), for 1/3-octave band labels.
• ISO 16:1975, Acoustics: Standard tuning frequency (A4 = 440 Hz).
• IEC 61672-1, Electroacoustics: Sound level meters, Part 1 (frequency and time weightings).
• OSHA, 29 CFR 1910.95 Occupational noise exposure, including Table G-16 and Appendix A.
• NIOSH (CDC), Noise and Hearing Loss: recommended exposure limit of 85 dBA, 3 dBA exchange rate, and workplace examples.
• NIDCD (NIH), Noise-Induced Hearing Loss (sound examples; 70 dBA and 85 dBA statements) and "How Loud Is Too Loud?" (NIH Publication 18-DC-6431).
• Kardous, C. A. and Shaw, P. B. (2014). Evaluation of smartphone sound measurement applications. J. Acoust. Soc. Am. 135(4), EL186–EL192; and the 2016 follow-up with external microphones, J. Acoust. Soc. Am. 140, EL327.
• Microsoft Learn, WAVEFORMATEXTENSIBLE structure and channel mask (multichannel channel order).
• ITU-R Recommendation BS.775, multichannel stereophonic sound system (5.1 reference loudspeaker layout).
• Microsoft Support, "Fix microphone problems" (Windows 11 steps).

Last reviewed 14 November 2026 by Sri Murugan Systems. Found a mistake? Tell us.