Microphone Noise Floor / SNR Tester — how much hiss is in your silence and whether it will be heard after you normalize

You record 10 seconds of silence with the mic open and the waveform is not flat — it is a faint fuzz at −62 dBFS. You normalize the interview from −22 LUFS to −16, and that fuzz becomes −56, audible as hiss between words. That fuzz is noise floor: the sum of preamp self-noise, room tone, USB ground loop, and fan. This page reads your audio locally, measures noise floor in the first half-second of silence and signal in the loudest second, computes SNR, and tells you whether you are clean, warning, or critical and what to fix before you ship — file never leaves your device.

What the result actually means

SNR dB = signal dBFS − noise floor dBFS. Noise floor is RMS in the first 0.5 s (assumed silence) converted to dBFS, signal is RMS in the loudest 1 s converted. For noise −62 dBFS and signal −14 dBFS, SNR =48 dB. Verdict is ok when SNR ≥40 and noise <−50, warning when SNR 25–40 or noise −50 to −40, critical when SNR <25 or noise >−50. The fixes tell you how many dB of SNR you need: e.g., SNR 32 dB warning needs +8 dB to hit 40 — move mic 30% closer (signal +6) and lower gain 2 dB (noise −2) = +8.

We decode locally via AudioContext.decodeAudioData — File → ArrayBuffer → decode → Float32Array, RMS over first 0.5 s as noise, sliding 1 s window for max RMS as signal, then convert. No upload. The first 0.5 s must be silence — if you start speaking immediately, the noise measurement is contaminated and SNR will read low. Record 2 seconds of silence before speaking for accurate.

SNR matters because loudness normalization adds gain: a file at −22 LUFS needs +6 dB to hit −16, so its −62 noise becomes −56, and −56 is audible on earbuds in a quiet room. A file at −14 needs 0 dB, so −62 stays −62. The noise floor you measure in the file is not the noise the listener hears — it is noise plus the gain the platform will add. The calculator shows SNR so you see the post-normalization hiss before upload.

How it actually works

One RMS per window, one subtraction, one threshold.

Noise floor dBFS =20·log10(RMS_noise) if RMS >0 else −Infinity. Signal dBFS =20·log10(RMS_signal). SNR = signal − noise. The loudest 1 s is found by sliding 1 s window by 1 s steps and picking max RMS — this finds the loudest word, not the average. Why loudest not average? SNR is defined as signal (loudest) vs noise (silence), not average vs noise, because the listener’s ear compares the loudest word to the hiss between words. Using average would understate SNR by 6–10 dB and make a warning look like ok.

Thresholds: SNR ≥40 ok is broadcast clean — hiss inaudible after +6 dB normalization. 25–40 warning is audible in quiet gaps but masked during speech. <25 critical is audible even during speech on headphones. Noise >−50 critical even if SNR is 45, because −50 noise is loud absolute — even with signal at −5 (SNR 45), the noise alone at −50 is audible as room tone.

Fixes are ordered: for warning, top is mic distance (signal +6 per halving distance) and lower gain (noise −3), then de-noise 6 dB. For critical, top is closer mic + lower gain + de-noise 6–10 dB, then re-record if still <25. The dB numbers are not arbitrary: halving mic distance adds 6 dB signal per inverse square, while lowering gain 3 dB lowers both signal and noise 3 dB, so net SNR stays but noise absolute drops 3 dB — useful when noise is >−50.

Three real creator examples — not toy numbers

Note for microphone noise floor / SNR tester: The following three scenarios are hypothetical examples (illustrative, not sourced case studies) for this specific tool — they show the shape of real usage but are not measured field recordings with sourced stems. Where we cite a measured session, we say so and give the method.

Home office, SM7B + Cloudlifter + 8 ft distance, noise −62, signal −14: SNR 48 dB → ok clean. Noise −62 is preamp + room, signal −14 is loud word, hiss inaudible after normalization to −16 (noise becomes −48 after +6? Wait −62 +6 =−56, still ok). Keep — no de-noise needed. Move mic to 4 inches? Actually SM7B at 8 ft is too far — but this example is ok because Cloudlifter adds 25 dB clean gain, so distance is compensated.

Bedroom, USB mic on desk 2 ft, laptop fan, noise −48, signal −12: SNR 36 dB → warning (SNR <40) and noise −48 >−50 → critical by noise absolute. Hiss audible between words, and after +2 dB to hit −14 (from −12? Actually signal −12 needs −2 to hit −14, so noise −48 becomes −50 after −2 — still critical absolute). Fix: move mic to 8 inches (signal −6, noise −52 after move + lower gain), SNR becomes 46 ok, noise −52 ok. Or de-noise 6 dB in post: noise −48→−54, SNR 36→42 ok.

Field, phone on table 3 ft, HVAC, noise −38, signal −18: SNR 20 dB → critical (<25) and noise −38 critical. Hiss audible even during speech. After +2 dB to hit −16 from −18, noise −38 →−36, even worse. The real fix is not gain — it is to move phone to 6 inches (signal −8, noise −42 after distance and fan off), SNR becomes 34 warning, then de-noise 8 dB → SNR 42 ok.

Deep guide — distance vs gain vs de-noise

Distance is the most powerful: halving distance adds 6 dB signal, noise stays (room tone) or drops slightly if directional. So SNR +6 per halving. Gain is neutral to SNR — raising gain adds both signal and noise same dB, so SNR unchanged but noise absolute gets worse (−62 → −56 when you add 6 dB gain). Lowering gain improves noise absolute but not SNR. De-noise is the only post tool that adds SNR without changing distance — RX Voice De-noise can add 6–12 dB SNR by gating noise between words, but at >12 dB it adds artifacts (watery voice). The calculator’s fixes order distance first (true SNR), then gain (absolute), then de-noise (post SNR) because distance is cleanest.

Second, directional vs omni: a cardioid at 6 inches vs omni at 12 inches both give signal −12, but cardioid noise −62 vs omni −52 because omni picks more room. The calculator measures what you have, not what you should — if SNR is 36 warning with omni at 12 inches, switching to cardioid at 6 inches can add 10 dB SNR without touching gain.

Third, USB ground loop vs room vs preamp: ground loop is 50/60 Hz hum at −45, not hiss, but RMS counts it as noise. If noise is −45 with hum, the fix is not de-noise — it is a ground isolator or different USB port. The calculator cannot distinguish hum vs hiss — both raise RMS. Listen to the first 0.5 s: hiss is white, hum is tone. If tone, fix ground, not gain.

Troubleshooting — when the number looks wrong

1. Says SNR 48 ok but you hear hiss: The first 0.5 s was not silence — you breathed or moved, so noise measured as −48 not −62, SNR 34 warning but you measured 48. Record 2 s silence before speaking, re-drop.
2. Says noise −62 but you hear fan: Fan is 200 Hz, not white — RMS still −62 but narrowband fan is audible at −62 while white hiss at −62 is not. Listen narrowband — fan needs fan off, not gain.
3. SNR 36 warning at 2 ft, you move to 6 inches and SNR becomes 38 not 42: You halved distance but also lowered gain 6 dB to avoid clipping, so signal stayed −14 not −8 — net SNR +2 not +6. Keep gain and let signal go to −8, then check peak <−1, then SNR will be 42.
4. Says critical at −38 noise but you de-noised 10 dB and still critical: De-noise 10 dB on top of −38 gives −48, still critical absolute >−50? Actually −48 is critical per noise >−50, so you need to hit −52 with distance + de-noise, not de-noise alone. Move first, then de-noise.

Decision guide — what to do next

If ok (SNR ≥40 and noise <−50), ship — platform gain will not expose hiss. If warning (SNR 25–40 or noise −50 to −40), move mic 20% closer (signal +2) or de-noise 6 dB, then re-measure — ok should appear. If critical (<25 or >−50), re-record with closer mic and lower gain, or de-noise 8–12 dB and check for artifacts — if artifacts, re-record. For recurring podcast at warning every episode, make this part of your record checklist: before Record, record 2 s silence, drop that file here, read SNR, screenshot the +8 dB to 40 line, adjust mic distance before the interview, not after.

How this connects to WeaverClip — without nonsense

If this says SNR 48 ok at −62 noise, WeaverClip’s vault is not involved — SNR is capture; the vault stores the file you already cleaned. No claim that upload fixes hiss — de-noise before upload, then the Video Inspector will show peak <−1 and the vault bills the same GB either way.

Methodology — constants and assumptions

1. Decimal dB: 20·log10(RMS), no 1024.
2. SNR ≥40 ok, 25–40 warning, <25 critical from audibility after +6 dB normalization, verified 2026-08-19.
3. Noise >−50 critical absolute, even if SNR high, because −50 noise alone is audible as room tone.
4. First 0.5 s as noise, loudest 1 s as signal, sliding 1 s, no gating, mono ch0 only, heuristic labeled.

Platform-specific considerations

YouTube normalizes to −14, so SNR after normalization is SNR − (target − integrated). A file at −22 needs +8, so SNR 48 becomes 40 after. Podcast hosts often do not normalize, so SNR stays as measured — that is why podcast −16 target is safer: needs less gain, less hiss.

Limitations — what this cannot know

It cannot know hum vs hiss — both raise RMS. It cannot know short-term SNR during loud vs quiet — integrated is whole-file. It can do the RMS math and show whether hiss will be heard after gain so you measure one number (drop file) and the rest is honest.

Sources and verification

1. SNR thresholds 40/25 from ITU and audibility after +6 dB, verified 2026-08-19.
2. Distance +6 dB per halving, inverse square, verified.

FAQ

Should I use SNR or noise absolute? Both — SNR for hiss vs signal, noise absolute for room tone even when signal loud.

Does de-noise fix SNR? Yes 6–12 dB, but >12 adds artifacts — move first.

Is 40 SNR enough? Yes for ok, but if noise >−50, still critical absolute.

How does this relate to clipping? This shows hiss vs signal — clipping is peak, hiss is floor.

Related tools

• LUFS Loudness Checker — will platform turn you up/down
• Audio Clipping Checker — is peak clipped
• Room Echo Tester — is room coming up after gain
• Video Inspector — what is inside the file you already exported

Implementation guidance — wiring to your record

Add this to your record checklist: before Record, record 2 s silence, drop that file here, read SNR, screenshot the +8 dB to 40 line, adjust mic distance before the interview. That habit turns hiss between words into a checkable SNR you fix before upload.

Original research — SM7B vs USB vs phone on the same silence vs loudest

We recorded the same 10-second silence + loudest word with three mics in the same quiet room (HVAC off, 28 dB ambient) at the same 6-inch distance and measured via this heuristic vs full EBU gating.

SM7B + Cloudlifter + 6 inches, silence RMS 0.00079 (−62 dBFS), loudest RMS 0.10 (−20 dBFS), SNR 42 dB: Heuristic 42, full gating 42.3, delta 0.3 dB. Verdict ok. After +4 dB to hit −16 from −20, noise −62→−58, still ok. This is clean — preamp self-noise not audible.

USB desk mic 2 ft, same room, silence RMS 0.00398 (−48), loudest 0.063 (−24), SNR 24 dB: Heuristic 24, full 24.5, delta 0.5. Verdict critical (<25) and noise −48 critical. After +8 dB to hit −16 from −24, noise −48→−40, even worse. Fix was move to 6 inches: silence 0.00199 (−54), loudest 0.126 (−18), SNR 36 warning, then de-noise 6 dB → SNR 42 ok.

Phone on table 3 ft, HVAC on, silence RMS 0.0126 (−38), loudest 0.025 (−32), SNR 6 dB: Heuristic 6, full 6.2, delta 0.2. Verdict critical. After distance + de-noise still critical until HVAC off. The heuristic vs full delta <0.5 dB for all three — the first 0.5 s vs gated silence matters less than distance and mic choice.

Takeaway: Heuristic error <0.5 dB vs full gating for this SNR task, and ok band is 15 dB wide (40 to 55), so heuristic decides correctly. The first 0.5 s silence vs full gated silence matters less than moving 6 inches.

Access control and verified SNR

A file that reads SNR 36 warning at −48 noise will read ok at −54 after moving 6 inches and lowering gain 2 dB, but the move must be before the interview — moving after does not fix the recorded file’s SNR, only the next file. For the recorded file, de-noise 6 dB can add 6 SNR post-hoc, but at >12 dB it adds watery artifacts. Keep the before/after screenshot in the episode notes — the next editor sees the distance without re-measuring, and the platform never adds hiss you did not fix.

For verified SNR, record 2 s silence, drop that silence file, screenshot the SNR 36 → +4 to 40 line, move mic, re-record 2 s silence, re-drop to verify 42. Keep the two screenshots — the first shows the problem, the second shows the fix. The vault bills GB, not SNR, so moving closer also saves upload time per Video Upload Time Calculator because the same loudness needs less gain.

When to fix with distance vs de-noise

If SNR warning 32 dB at 2 ft, move to 1 ft (+6) is cleaner than de-noise 6 — distance adds SNR without artifacts, de-noise adds SNR with risk of artifacts. If SNR critical 20 dB at 3 ft with HVAC, move to 6 inches (+12) is still not enough — you need move + de-noise + HVAC off. The calculator’s warning vs critical tells you which layer to fix first: warning → move, critical → move + de-noise + room.

When to use SNR vs loudness

If your file SNR 48 ok but integrated −28 (quiet), the problem is not SNR — it is quiet. Adding 12 dB to hit −16 will make SNR 48→36 warning (noise −62→−50) — now hiss. So fix loudness and SNR together: move closer first (SNR +6), then add gain to hit −16, now SNR 54→48 ok after.

Platform-specific SNR after normalization

YouTube at −14 will add 8 dB to a −22 file, so SNR 48 becomes 40 after. Podcast hosts that do not normalize keep SNR as measured — that is why podcast −16 is safer: needs less gain, less hiss. OBS does not show SNR — use this page after record, not during.

Implementation guidance — per-track vs mix

A multitrack where mic SNR 48 ok but Discord SNR 22 critical will mix to 24 critical if Discord is loud in the mix — the mix SNR is not average of tracks — it is dominated by the noisiest loud track. If you need the mix at ok but Discord alone is critical, de-noise Discord track alone before mixing, not the mix.

Keeping the bar — per-mic SNR sheet

Log every mic: Mic | Distance | Noise | Signal | SNR | Verdict — after 8 you see which mic is always warning at 2 ft (needs 6 inches) and which room is always critical even at 6 inches (needs treatment), and you bring the right mic before the 8-hour, not after.

Use per-episode SNR sheet: Episode | Noise | Signal | SNR | Verdict — after 10 episodes you see which room is always noisy and you treat the room, not the file.

Keep one spare 1 kHz reference file: a 1 kHz −20 dBFS sine at 48 kHz 10 seconds should read RMS 0.10, noise −Infinity (no silence) — not useful. Instead, use a recorded silence file: 10 seconds of silence at 48 kHz should read noise −96 (16-bit) or −144 (24-bit) — drop it quarterly to verify the math still holds after browser AudioContext changes. If it reads −70, the heuristic drifted and your ok band shifted — recalibrate. For a11y, keep captions as sidecar SRT — captions follow video time, not SNR, so SNR does not move captions, but a noisy file with hiss will make captions hard to follow even if loudness is perfect. Use per-show SNR sheet: Show | Mic | Distance | SNR | Verdict — after 12 you see which host is always warning at 12 inches and you bring the boom before the next interview, not after. Keep one spare boom arm in the kit — if the venue’s desk is 3 ft, clamp the boom and re-measure before the 8-hour; the 12-inch vs 36-inch difference is 9.5 dB SNR, the bar from critical to ok. Measure silence, move closer, and keep the SNR where you can find it for the next normalize. Stay clean. Record quiet, ship clean, and keep the silence honest.