What Is the Audio Noise Floor?

The noise floor is the time-averaged level of background noise in the absence of intentional signal, measured in dBFS (decibels relative to digital full scale; a lower / more-negative number is quieter). Two common variants: peak-noise-floor (the loudest sample in a 'silent' section) and RMS-noise-floor (the average power over a measurement window of 400 ms typically). Voice-over QC uses peak; mastering uses RMS. The noise itself is the sum of many uncorrelated sources — thermal noise in the microphone capsule and preamp, quantisation noise of the ADC, room HVAC, computer fan, traffic, electrical mains hum — that combine into a roughly white or pink hiss with occasional tonal contributions. See [what-is-audio-bitrate](/guide/what-is-audio-bitrate) for related dB scales.

Targets differ by delivery context. ACX audiobook submission: peak noise floor at or below -60 dBFS — non-negotiable, automatic rejection above this. Broadcast voice (BBC, NPR, EBU): -60 to -65 dBFS depending on programme type. Podcast: below -50 dBFS is industry-acceptable; below -55 dBFS is a 'quiet' show. E-learning and corporate VO: typically below -55 dBFS. Music recording: dynamic range demands a -70 to -80 dBFS floor on a quiet ambient room mic. Live broadcast (sports, news): -50 to -55 dBFS achievable with directional mics in noisy environments. Field recording (documentary): -45 dBFS is often the best achievable in real environments; post-production noise reduction must close the gap. See [audio-for-voice-actors](/guide/audio-for-voice-actors) and [audio-for-podcasters](/guide/audio-for-podcasters).

Quantisation noise sets the absolute floor a converter can deliver. 16-bit linear PCM: -96 dBFS. 24-bit: -144 dBFS — well below any analogue source. Dithered 16-bit can reach an effective noise floor of about -120 dBFS using noise-shaped dither. In practice, the limiting factor is never the converter but the analogue chain: a Neumann TLM 103 self-noise spec is 7 dBA equivalent SPL — at typical VO close-mic gain that translates to about -75 dBFS noise floor. A budget USB condenser sits at 17-22 dBA, translating to -55 to -60 dBFS. The room contributes 30-50 dBA SPL of HVAC and ambient sound; through a -60 dB mic gain stage, that adds another -30 to -35 dBFS of noise. The recorded floor is dominated by whichever is loudest of these three.

Method: capture 5-10 seconds of room tone (mic running, no talking, normal session conditions), then analyse. In Audacity: Effect > Analyze > Sound Finder shows the peak; or select silence and run Effect > Amplify — the dialog reports the peak amplitude. Convert lossy sources to [WAV](/convert/mp3-to-wav) before measurement; lossy compression hides low-level noise. Adobe Audition: Window > Amplitude Statistics gives RMS and peak. iZotope Insight 2: shows continuous LUFS-S and peak meters; the lowest LUFS-S value during a silent passage approximates the integrated noise floor. ACX-specific verification: Audacity's free 'ACX Check' plugin reports peak noise floor, RMS, and peak in a single pass. Always measure with the same gain, mic, and room conditions you used during the take — measuring later with different gain gives a different number.

Source-level reduction always beats post-production processing. (1) Mic choice: pick the lowest-self-noise microphone you can afford — for VO, large-diaphragm condensers from 7 to 12 dBA self-noise (Neumann TLM 103, Sennheiser MK4, AT4040). (2) Room treatment: treat reflections with absorbers and diffusers; isolate the recording space from HVAC and traffic. A vocal booth or PVC-frame portable booth (Aston Halo, sE Reflexion Filter) drops floor 3-6 dB. (3) Preamp gain staging: set gain so peaks land at -12 to -6 dBFS; running the preamp lower with a hot makeup stage adds noise. (4) Quiet hardware: laptop fans, desktop drives, and projector fans all radiate noise — kill them or move them out of the room. (5) Power: use a surge-protected linear supply; switch off LED dimmers; see [fix-audio-hum](/guide/fix-audio-hum) for related interference paths.

When the source-level floor is fixed and you need to drop another 6-15 dB, post-production noise reduction is the next step. iZotope RX 11 Voice De-noise and Spectral De-noise are the industry standard; a 6 dB reduction is transparent on speech, 12 dB introduces faint artefacts, more than 15 dB starts to sound 'underwater'. Adobe Audition: Effects > Noise Reduction / Restoration > Noise Reduction (Process) — capture a profile from a silent section, set Reduce by 6-12 dB. Free path: Audacity Effect > Noise Reduction with the same workflow (capture profile, then apply with Reduction 6-12 dB, Sensitivity 6, Frequency Smoothing 3). Apply noise reduction before EQ and compression; downstream gain stages would re-amplify residual noise. Always A/B against the original — over-reduction is more damaging than the noise itself.

Lossy codecs alter the noise floor. MP3 and AAC encoders apply psychoacoustic masking that often raises the digital noise floor above the source's analogue floor — the encoded file may show a -50 dBFS floor when the source WAV measured -65 dBFS. This is normal and inaudible if the bitrate is high enough (192 kbps stereo or above). Below 96 kbps, the codec's quantisation noise dominates and may itself exceed delivery spec floors; do not encode VO at 64 kbps. When converting an archive, [WAV to FLAC](/convert/wav-to-flac) preserves the noise floor exactly because FLAC is lossless. [WAV to MP3](/convert/wav-to-mp3) at 192 kbps preserves perceptually; below that, the encoded floor changes. Always master to spec and then encode — never spec-check after lossy encoding.