Audio Mastering on Your Phone: What the Chain Actually Does | WaveCraft

The short answer

Mastering is the last pass that makes a mix hold together on someone else's phone speaker: a consistent level, controlled dynamics, and a peak that never crosses into clipping. In WaveCraft it is not a five-knob black box and it is not an EQ-and-stereo-widening stack — it is a compressor, a limiter, and a loudness target, chosen by where the track is going. Three presets cover the whole job, each a set of real numbers you can read before you export.

The honest part matters just as much: mastering finishes a mix, it does not repair one. It cannot pull a buried vocal forward, un-clip a recording, or turn a rough mix into a record.

WaveCraft on the web describes the app; it is on the App Store for iPhone and iPad on iOS 17+.

Why you are here

Most people asking this are not looking for a mastering course. They are stuck at the same moment: the track is finished, it sounds almost right, and something is missing that they cannot name. Or they are about to upload something and want to know whether a phone can do the step at all.

A phone can do mastering — the operations are ordinary DSP, and screen size is irrelevant. What determines the result is whether the mix underneath is finished, and whether the level suits the destination.

Wasted effort usually goes on a second misunderstanding: the generic description of mastering — EQ, compression, stereo widening, limiting — describes a chain that in WaveCraft is not on the master at all. EQ, reverb, chorus, phaser, delay, distortion and de-essing are effects that live on tracks and clips, shaping individual parts. The master bus is only the last two dynamics stages plus a level. Brightness, air and width are track decisions.

What the master bus actually does

The compressor evens out the loud and quiet parts so the track holds its level. How little you need is the point: a mix bus wants a gentle ratio and small reduction — the Streaming preset runs at 2.5:1 above −12 dB. Heavy compression is a creative decision, not a finishing step.

The limiter holds the peak below a hard ceiling so nothing clips. It attacks instantly and releases slowly, because a fast release turns a sustained passage into audible chatter. In WaveCraft it is not a separate module: compression and limiting are two halves of one dynamics engine, and the ceiling is a hard clamp at the end of the chain.

The loudness target is the final level, and the part people misunderstand — so it gets its own section.

That is the entire chain. No hidden spectral stage, no exciter, no width.

The three presets, in real numbers

Each preset is a destination, not a ratio. The three ship with fixed settings:

Read across a row and the intent is obvious. Podcast runs quieter than streaming but compresses twice as hard with half the attack time, because speech needs consonants caught at the front of each syllable. Loud is the opposite: the fastest attack and the tightest ceiling in the set, so it hits hard without crossing zero.

Selecting any preset takes over normalization entirely — the separate loudness toggle disappears, because a preset that carries its own target and a toggle that sets another one is a contradiction. With no preset selected, the toggle targets −14 LUFS on its own.

The values are not secret and they are not approximate: the app shows the target number for whichever preset you pick.

Loudness target versus peak ceiling

This is the mechanic worth understanding, and it explains why the presets are safe rather than reckless. To hit a loudness target the exporter measures the finished master, computes one gain, and applies it. But pushing a track up by gain can push its peak over the ceiling, and a clipped master is worse than a quiet one. So the gain is not applied blindly. The exporter takes the smaller of two: the gain that would reach the loudness target, and the gain that would just bring the peak down to the ceiling.

Three consequences follow:

• The loudness target is a target, not a guarantee. A track whose peaks are already huge relative to its average will land below the number you chose. You get an unclip-safe file instead of a louder one, every time.
• The ceiling is absolute. Whichever preset is selected, the peak never crosses it. Staying below 0 dBFS also avoids inter-sample clipping, which causes the flat, brittle distortion on good speakers.
• You cannot out-shout the target by asking harder. "Loud" at −9 LUFS is the most aggressive setting that keeps its promise. Past it you are making a mastering choice by hand, with a computer and a meter.

With no preset selected, normalization still refuses to push a file into clipping.

What the meter measures

The loudness figure comes from a real meter, not an estimate. WaveCraft measures ITU-R BS.1770-4 loudness, the same standard streaming services use: K-weighted frequency shaping, read in 400-millisecond blocks with 75% overlap, so a new block arrives every 100 milliseconds.

The gating is what makes it usable. Blocks below −70 LUFS are excluded outright, and a relative gate drops anything more than 10 LU under the track's own level — without that second gate a quiet intro would drag the average of the whole song downward. Export normalization uses the integrated figure, the gated average over the entire file, not the momentary or short-term reading.

One ordering detail is easy to get wrong: the master chain runs before the measurement. The engine renders the mix once with the chain already applied and measures that. Measuring the pre-master mix would mean the meter describes audio that does not exist in the file you ship.

Two edge cases, because they surprise:

• A track too short or entirely silent to measure is exported at unity gain rather than scaled by a guess. A three-second stab does not get amplified into noise because of it.
• Stem exports share one gain. Normalizing each stem separately would erase the relative levels between them — the exact opposite of what you want when exporting tracks to remix elsewhere. All stems get the same gain, so their balance survives.

What mastering cannot fix

The useful half of understanding mastering is knowing where it stops.

It cannot fix a bad mix. A vocal buried under the instruments stays buried; no amount of compression moves it to where it belongs.

It cannot restore a clipped recording. If the waveform was captured flat-topped, that information was never recorded. A limiter cannot reconstruct it — only refuse to make it worse.

It cannot fix a bad performance. Timing, tuning and arrangement are upstream decisions.

It cannot make a demo sound like a studio recording. It makes a good mix sound finished, and finished on every playback system. That is the honest scope.

What WaveCraft does differently

• The master bus and the effects panel run the same dynamics engine. A 4:1 compression in the effects panel and a 4:1 on the master are literally the same code — no second implementation that sounds subtly different, and no Apple dynamics unit whose knobs quietly do not do what they say.
• No look-ahead, so nothing shifts. The chain introduces zero latency, so a track with effects never drifts against its neighbours on the timeline. The cost is that a transient's first sample can momentarily exceed the ceiling; a hard clamp at the end makes "the peak never crosses the ceiling" a guarantee rather than an approximation.
• Stereo-linked detection. Peak is the maximum across both channels per frame, so the image does not sway as one side gets pulled down and the other does not.
• State that survives buffer boundaries. Gain is continuous across render buffers rather than resetting at each one, so quiet passages have no gain steps listening as clicks.
• Everything local. Measurement, chain and export run on your device — never uploaded, no account, works offline. Pro unlocks every mastering preset, the 20+ effects, unlimited AI separation and denoise, multi-track mixing and lossless WAV, ALAC and FLAC up to 32-bit float: monthly, yearly, or a one-time purchase with Family Sharing, with a 7-day trial on the subscription plans.

Before you export

1. Is the mix finished? A buried part, a sloppy take, or a clipped track is track-level work. Fix it on the timeline before touching the master.
2. Pick by destination. Streaming for anything going to a streaming platform, Podcast for speech, Loud only when you want the compression that comes with it. If you cannot say why, Streaming is the safe answer.
3. Export lossless. WAV, ALAC or FLAC, up to 32-bit float. AAC is there for delivery where size matters; there is deliberately no MP3 option, because iOS has no native MP3 encoder and the app does not offer a format it cannot guarantee.

Then listen on headphones and on a proper system before you call it finished. That limit is not about the phone — it is about judging a master on a two-inch speaker, which is true on any device.

Download WaveCraft on the App Store — the master chain is the last two minutes of the workflow, and the rest is free to explore first.