What is microphone noise floor in dBFS?
Noise floor is the level of unwanted background signal — room tone, electrical hiss, computer fan, distant traffic — that your mic captures when nobody is speaking. dBFS (decibels relative to full scale) is the digital reference where 0 dBFS is the absolute loudest sample possible and everything quieter is negative. A great noise floor is below -60 dBFS, broadcast quality is below -55 dBFS, acceptable streaming is below -50 dBFS, and anything noisier than -45 dBFS will be obvious to listeners as 'fuzzy' background.
What is a good target for streaming or podcasting?
Streamers and Twitch broadcasters can usually get away with -50 dBFS or quieter — game audio and music will mask the floor. Podcasters and audiobook narrators should aim for -60 dBFS or below, because long-form spoken content has lots of pauses where the floor becomes audible. Music studios target -70 dBFS or quieter, often using treated rooms and outboard preamps. If your floor is louder than -45 dBFS, work on the room (close windows, mute the PC fan, move the mic) before buying a new mic. Compare against the LUFS spec at https://devicecheckonline.com/microphone/lufs.
Why is my noise floor so high even with a quiet room?
Common causes, in order: input gain too high (every +6 dB doubles the floor too), USB power noise from the same hub as the mic, ground loop from an unbalanced cable, fluorescent or LED light driver noise, electrical hum from a nearby transformer, fan or hard-drive noise picked up acoustically. Try unplugging everything except the mic, lowering gain, and using a different USB port directly on the motherboard. A condenser mic in a reflective room can also pick up reverb tail that registers as noise.
Why are median, p95, and peak reported separately?
Because noise is rarely constant. The median (50th percentile) is your true sustained floor — what you hear during silence. The p95 catches occasional spikes — a key press, breath, footstep — that the median hides. Peak is the single loudest sample, useful for detecting transients you might not consciously notice. A median of -65 dBFS but p95 of -40 dBFS means the room is mostly quiet but full of small clicks — fix the clicks, not the gain. Watch live levels at https://devicecheckonline.com/microphone/waveform.
Does browser noise suppression affect this measurement?
No — we explicitly request `noiseSuppression: false` and `autoGainControl: false` via getUserMedia, so Chrome and Edge serve the raw signal the OS hands them. This is critical: with suppression on, browsers report an artificially clean -70 dBFS even from terrible mics, because the algorithm is gating out everything quieter than speech. Apps like Zoom and Discord apply their own suppression on top. To measure the real hardware, every test must request the raw stream as we do here. Verify your gate threshold separately at https://devicecheckonline.com/microphone/gate.
How do I lower my noise floor?
Cheap wins first: lower input gain and move closer to the mic (every halving of distance gains 6 dB of signal without adding noise), point the mic away from PC fans and windows, switch to a dynamic mic if the room is untreated. Medium wins: shock mount, pop filter, foam panels behind your seat to kill reflections. Big wins: a dedicated audio interface instead of USB mic, a higher-end preamp, and acoustic treatment. AI tools like RTX Voice and Krisp can mask remaining noise in real time but should not be a substitute for a clean signal.