The Five-Step Audio Pipeline: Trim, Denoise, Loudness, Fade
Raw recordings need only a handful of operations, but wrong order makes them fight each other: loudness before denoise amplifies the noise floor; fading before trimming cuts the fade off. The sequence is the whole craft.
Step one: trim — fix the range
Cut the usable span first so every later step acts only on it. Trimming shrinks the battlefield: dead openings, table-banging endings, long silences — gone before processing, making denoise faster and loudness statistics cleaner.
Endpoint technique: start 0.3 seconds early (leave a sliver of room before the first syllable or it sounds muffled), end 0.5 seconds late (let natural decay finish). Listen to both ends immediately after cutting.
Steps two and three: denoise, then loudness
Denoise before loudness — never reversed: denoise engines separate signal from noise by contrast, and boosting loudness first raises the noise floor with it, blurring the very features the engine needs. Voice material picks the AI engine (RNNoise, voice-optimized); general material picks spectral denoising. Start at standard strength — light for clean rooms, strong for noisy venues, accepting that higher strength costs voice detail.
Then normalize loudness to a consistent target across the clip. It must follow trim (range fixed) and denoise (noise no longer skews the statistics) — that is what makes "consistent" meaningful.
Steps four and five: fade, then export
Fades come after loudness: loudness moves the whole envelope and would reshape any fade applied earlier. Now finish: 0.3s fade-in to clean the attack, 2s fade-out to close naturally, click removal at any splice points.
Export last: WAV for further editing in a DAW (maximum headroom); MP3 high quality for direct publishing (small, universally compatible). Five steps, usually under five minutes total — and the result sounds like a different recording than the raw file.