LevelUp AudioTwo recordings can have the same loudest moment while one still sounds consistently quieter. A recording with one brief scream reaching its peak and a recording with consistently clear speech carry different average levels of sound.
Peak and perceived loudness answer different questions
Peak is the highest instantaneous level. LUFS summarizes loudness over time using characteristics of human hearing. Applying the same gain to an entire file makes everything louder or quieter, but leaves the ratio between whispers and loud sounds within that file unchanged.
How LevelUp Audio processes sound
In Microphone voice mode, the tool examines levels over time and raises quieter sections more. A compressor gently reduces louder moments. A limiter controls high peaks, then measured values are used to balance the overall loudness. The tool does not delete specific words or trim silent sections.
The final targets are −16 LUFS average loudness and −1.5 dBTP true peak. These are this tool's targets, not mandatory standards for every platform. Results may differ depending on the recording, so check the displayed values. At 0% strength, no leveling is applied.
Where natural sound is easily lost
Boosting very quiet speech can also raise nearby fan noise and breaths. Strong leveling can reduce a game's tension or a speaker's expression by flattening dynamics too much. Mixed mode uses gentler volume changes to retain more of the game's atmosphere, but cannot selectively process the voice alone.
A fair comparison is not simply a louder one
Match playback volume between the original and the result, then listen for quiet word endings, comfortable loud moments, and swelling background sound. Word endings already missing from the original cannot be recovered by increasing strength. Check your recording filters or microphone input first in that case.
What testing has and has not established
On 2026-10-11, the updated release at the default 60% strength reduced the segment volume gap from 27.2dB to 1.8dB in a real English speech recording with artificially varied levels. Duration was unchanged and no clipped samples were detected. This does not verify subjective quality for Korean whispers and screams or every mixed game recording. Compare your own source and result.
Technical references: FFmpeg dynamic audio normalization, FFmpeg loudness normalization. Product descriptions follow the actual processing order in the deployed code.