Voice & Vocal Enhancement — AI Speech Cleanup

Upload the recording — the tool recognizes speech and switches AI noise reduction on. It targets steady background: fridge or fan hum, street noise, air conditioning, microphone hiss, dishes. Other people’s voices and a TV in the room are speech too, so they get quieter but do not disappear entirely. When there is a lot of background, pick the Phone Call preset and raise Enhancement Strength.

Voice enhancement is specifically optimized for speech and vocal recordings. It includes voice-specific modules like de-reverb, de-essing, and de-click that target common speech recording problems. The EQ is tuned to voice frequencies (200 Hz – 10 kHz), and loudness targets follow broadcast standards (LUFS).

LUFS (Loudness Units Full Scale) is the broadcast standard for measuring perceived loudness. Spotify uses -14 LUFS, podcasts typically use -16 LUFS, and broadcast TV uses -23 LUFS. Normalizing to these standards ensures your voice sounds consistent and professional across all platforms. For vocals and music, loudness is raised only when the recording is quieter than the target level — a loud master stays untouched.

Yes. When you upload singing, the tool detects a solo vocal and switches to the Vocals mode: no de-esser or gate, soft compression, and a careful EQ — the timbre and breaths are preserved. The Enhancement Strength slider lets you bring the result even closer to the original, and the "Source" quality setting returns a file with the same bitrate as the source.

Partly, and it helps to know which part. Boxiness is the room, not the microphone: a bare wall behind you sends the same voice back a few milliseconds later, and that reflection is recorded at almost the level of the voice itself. Narrow cuts in the 200–400 Hz region take the worst of it out and the enhancer does that automatically. What it cannot do is separate the reflection from the original when both are equally loud — so the real fix is on the next take: get closer to the microphone and put something soft, a curtain or a coat, on the wall you face.

Automatic first, always. The tool listens to the file, decides whether it is speech, singing or music, and picks one of seven presets accordingly — Podcast, Interview, Voiceover, Vocals, Phone Call, Old Recording or Music. That is right often enough that manual is the exception rather than the route. Switch to manual when the automatic choice is audibly wrong for the material, or when one module is doing damage: there you can turn each of the nine on or off and hear the difference one at a time.

They are two noise measurements, not a score. "Background in pauses" is the level of the quietest parts of the recording, between words, in dB below full scale. "Background vs. voice" is how many decibels quieter than the voice it is. The bands follow published norms: under 40 dB the background is noticeable (the ACX audiobook standard asks for about 40), from 40 to 50 dB it is audible in pauses, especially on headphones, and from 50 dB it is not heard — that is how voice-over studios record. After processing the same readings show the result, and the "dB voice-to-noise" line shows how far the background went down. For music and singing the background is not measured: their quietest parts are the music itself.