Know what you are measuring.

The capture path

The browser requests your microphone after you press Start. An AudioWorklet receives audio samples for a guided twelve-second capture: three seconds of room tone followed by nine seconds of speech. The tool produces mono playback, a waveform with a quiet/speech marker, segment listening, and a WAV download. There is no live playback loop or audio upload.

We analyze the captured samples, not the pixels in the display or a device price. Capture hardware, operating-system processing and browser processing can all affect those samples. The receipt includes settings the browser reports, with unavailable values treated as unknown.

RMS and dBFS

Root mean square (RMS) describes the average magnitude of the samples in a section. For normalized samples x, the calculation is:

RMS = √(sum of x² ÷ sample count)
level in dBFS = 20 × log₁₀(RMS)

Digital full scale is the reference. A normalized amplitude of 1 is 0 dBFS; 0.1 is −20 dBFS; 0.01 is −40 dBFS. These are mathematical examples. Exact digital silence has no finite logarithmic level and is shown as unavailable rather than presented as a measured negative infinity.

Room tone and speech are separate sections

The opening section measures the background captured while you are asked to stay quiet. The later section measures whatever you record during the speaking prompt, including pauses. MikeNet does not use voice recognition to verify that you spoke. A silent or incomplete take cannot establish speech quality.

Speech level minus room-tone level is a difference between two sections recorded at different times. It is not a laboratory signal-to-noise ratio or an intelligibility score. A changing fan, an unexpected sound or automatic gain can alter the comparison. You can play each section alone to hear what the numbers summarize.

Peak, crest factor and near-full-scale samples

Peak uses the largest absolute sample value, expressed in dBFS. Crest factor is peak dBFS minus RMS dBFS for the full take. A higher crest factor means a larger gap between peak and average digital level — useful dynamics evidence when comparing takes, not a grade for the microphone.

Samples close to full scale are worth listening to carefully, but their count does not establish where distortion occurred. A microphone or interface can distort before the browser receives the signal; a limiter can prevent full-scale peaks without undoing that distortion.

We count samples whose absolute amplitude is at least 0.99 of digital full scale. This threshold is shown with the result. The statistic is a warning to inspect the audio, not a hardware-failure verdict.

Comparing A and B

Read the same prompt with a similar voice and placement unless placement is the one thing you are testing. Listen at the same playback volume on headphones. Use Flip A ↔ B to keep the playhead while switching takes. Name the change so the receipt has context. Different devices or processing settings can make a numerical comparison less direct; those differences must stay visible.

For example, room tone and speech both falling by 10 dB is a quieter overall capture, not evidence that their separation improved. If only room tone falls, listen for gating or damaged word beginnings before calling it a useful improvement.

Speaker playback mixes the room back into the judgment. Prefer headphones and stop playback before the next capture so the next take does not record the previous one. See the headphones guide and the echo guide.

Browser constraints are requests

The option to request processing off asks the browser to disable echo cancellation, noise suppression and automatic gain control. A browser may not support every request, and the operating system or device can still process audio. We show the settings actually reported by getSettings(); we do not infer missing values.

What remains outside this test

• Calibrated dB SPL, hearing-safety measurements and microphone sensitivity.
• A certification of frequency response, latency, electrical isolation or hardware health.
• Exactly how another app, codec, network connection or listener’s speakers will reproduce your voice.
• A universal score for microphone or room quality.

Reproducible examples

The tool’s demonstration uses generated signals so you can inspect the workflow without microphone access. Demo audio is explicitly labelled and is not presented as a person, a real room or a tested product. Downloaded receipts describe the source of each take.