Noise Floor
The noise floor is the level of constant background sound present in a recording when nobody is speaking. Room hum, computer fans, traffic through a window, the self-noise of the microphone and preamp: everything that remains when the narration stops. ACX requires it at -60 dBFS or lower.
Sixty decibels below full scale is quiet. It is roughly the difference between a normal speaking voice and a sound you would struggle to notice in a silent room. It is a demanding target for home recording and a routine one for a treated space.
Why distributors care about a sound nobody notices
A listener does not consciously hear a high noise floor on any single sentence. They hear it across eight hours, as fatigue. It also survives compression badly: the same processing that evens out narration lifts the quiet parts, so a noise floor that was borderline in the raw recording becomes audible hiss in the finished file. Gaps between sentences are where it shows first.
What raises it
- The room. Hard parallel surfaces reflect sound back into the microphone. Most domestic rooms are worse than their occupants think.
- Air handling. Heating and air conditioning produce broadband noise that sits exactly where narration lives.
- Gain staging. Recording too quietly and raising the level afterwards raises the noise with it. The noise floor is set at capture and cannot be lowered later.
- Cheap preamps. Self-noise from the interface itself puts a hard limit on how quiet the recording can be, no matter how good the room.
Why it cannot be fixed in post
Noise reduction software removes noise by subtracting a profile of it from the whole signal, and the voice occupies the same frequencies as the noise. Aggressive reduction takes the noise out and leaves the narration sounding thin, watery or metallic, a texture usually described as artefacting. Reviewers who reject audio for over-processing are hearing the cure, not the disease. The only reliable fix is to record quieter in the first place.
Room tone is not the noise floor
Related and often confused. Room tone is a deliberate half to one second of recorded silence placed at the head and tail of every file, so playback does not cut abruptly from nothing to speech. The noise floor is the level of that silence. ACX requires both: room tone present, and quiet enough.
Where AI narration differs
Synthesised narration has no room and no microphone, so its noise floor is a function of the rendering and mastering chain rather than of a physical space. That removes the hardest problem in home recording, and it is why the -60 dBFS requirement is rarely the constraint on an AI-produced file.