Audio compressor — tune threshold / ratio / attack / release and export WAV

Compression happens at the stage when the recording is most sensitive

You reach for a compressor at a particular moment: the track is almost finished but not yet published. Mastering a song, polishing a podcast episode, levelling a voice-over for a client. At this stage the file contains everything — the performer’s voice, the musical arrangement, lyrics, the sonic character that defines the work. It hasn’t been released yet and the rights may not be fully settled.

Dynamics processing requires the original, uncompressed audio to work on. An online tool that applies a DynamicsCompressor on its server has to receive the full audio file first. The technical necessity is real; it just means that ‘trying a compressor setting’ and ‘handing over the final mix candidate’ become the same action.

How cloud audio services handle the files you send

Server-side audio processors receive your file, process it, and return the result. In the free tier, terms of service commonly grant the service rights to use uploaded content — for feature development, for training, for other purposes described in language that is hard to parse at the time of upload.

Each parameter tweak that triggers a re-process sends another request. Over a session spent dialling in compression settings, the original file may have been transmitted multiple times. None of this is inherently malicious — but the file is there, on infrastructure the user doesn’t control, at the most sensitive point in the production timeline.

DynamicsCompressorNode processes everything inside your browser

The tool decodes audio via decodeAudioData, chains a DynamicsCompressorNode (threshold, ratio, attack, release, knee) followed by a GainNode for makeup gain, renders the result with OfflineAudioContext, and writes 16-bit PCM WAV. Nothing crosses the network; re-renders on parameter change run entirely in browser memory.

Open DevTools Network while adjusting parameters: no audio requests appear. The source — including the DynamicsCompressorNode configuration and the offline render path — is on GitHub for review.

Things worth checking before and after compressing

Compression is destructive. The original dynamics can’t be recovered from the processed output, so save the WAV alongside your source file rather than over it. Start with a preset, listen carefully, and adjust threshold and ratio to taste — that feedback loop runs in the browser without any uploads.

Makeup gain is a separate concern from the compression itself. Applying too much lifts the average level beyond what the downstream limiter or final encoder expects, potentially causing clipping. Pair this tool with audio-loudness-lufs and audio-true-peak to check levels before committing to a final export. If you only need a static gain change rather than dynamics processing, audio-volume does that without altering the envelope.

Threshold, ratio, knee — what the parameters actually do

A compressor watches the input level and, whenever it exceeds the threshold, divides the overshoot by the ratio. With threshold = -18 dB and ratio = 4:1, a -10 dB input (8 dB above threshold) is reduced to -18 + (8 / 4) = -16 dB at the output. A ratio of 1:1 passes signal through unchanged; ∞:1 becomes a brick-wall limiter. Knee controls how the transition around the threshold is shaped: a hard knee (0 dB) compresses abruptly at the threshold, while a soft knee (20–40 dB) eases into compression — preferred for vocals and acoustic guitar where the engagement should be invisible to the listener.

Attack is the time it takes for gain reduction to reach its target after the signal crosses the threshold; release is how long the gain reduction takes to recover once the signal drops back. The Web Audio DynamicsCompressorNode runs an internal RMS-style envelope follower with exponential time constants set by attackTime and releaseTime. An attack of 3 ms clamps down on transient kicks and consonants; 30 ms lets the transients through and only manages the sustained energy. Release in the 100–300 ms range avoids audible pumping (level breathing) — drum-dense material may want 60–80 ms, while voice and dialogue start around 150–250 ms.

Genre-typical settings and what compression actually fixes

Voice / narration / podcast work commonly starts at threshold -18 to -14 dB / ratio 3:1 to 4:1 / attack 10–20 ms / release 150–250 ms — controlling sibilance and plosives while lifting the average to improve intelligibility. Mastering glue compression on the mix bus uses threshold -8 to -6 dB / ratio 1.5:1 to 2:1 / attack 30 ms / release 100 ms to bind the mix together. Drum-bus compression with threshold -10 dB / ratio 4:1 / attack 5–10 ms / release 50–100 ms preserves attack and brings up the sustained energy.

Common pitfalls: (a) pushing ratio too high flattens musical dynamics and removes the sense of breath, (b) very short attack times kill the consonant and drum transients, leaving things sounding dull, (c) long release times raise the decay tails of piano and acoustic guitar in an unnatural way, (d) heavy makeup gain pushes the integrated loudness past streaming targets, causing platform-side auto-attenuation (Spotify / Apple Music / YouTube at -14 LUFS, Apple Podcasts at -16 LUFS, EBU R128 broadcast at -23 LUFS). Measuring with audio-loudness-lufs and audio-true-peak as a final pass keeps the output in spec for whichever platform you’re targeting.