What Is Audio Mastering? The Signal Chain Breakdown
Summary
Audio mastering is the last step between your mix and the version that goes to Spotify, Bandcamp, or a club PA. It corrects tonal imbalance, controls dynamics, and sets loudness to platform targets. Most streaming platforms normalize to -14 LUFS. AI mastering tools handle the job for most releases. This breaks down the signal chain and explains when automated mastering holds and when it doesn't.
Audio mastering is what happens to your track between the final mix and the streaming platform. It's the last technical step: a corrective and creative pass on the stereo output that sets loudness, tonal balance, and dynamics for every playback environment the music will hit. One session. Every system after that is out of your hands.
Most producers understand mixing. What is audio mastering, specifically, stays blurry for a lot of people who've been producing for years. This breaks it down without padding.
Audio mastering is the last filter before your track goes public
Your mix is finished. Every element sounds right in the room. But a finished mix is not a finished master. The mix is optimized for one context: your DAW, your monitors, your room. Mastering prepares the audio for every context: earbuds at -14 LUFS on Spotify, a club sub at 110 dB SPL, a vinyl lathe at half the dynamic range.
The mastering engineer (or the mastering chain you set up yourself) works on the stereo output, not the individual tracks. That's the constraint. You can't fix the kick or pull down the vocal. You're working on the sum of everything that happened in the mix.
That's why a good mix matters before mastering. Mastering corrects problems at the macro level. It can't undo a buried snare or a phase-cancelled bass. What it can do is translate a solid mix onto any system that plays it.
What the mastering chain actually does to your signal
Three main tools: EQ, compression, and limiting. In that order, usually.
EQ adjusts the frequency balance of the full mix. If the low end is too heavy or the high mids are brittle, corrective EQ fixes it. Creative EQ adds presence, warmth, or air. Most mastering sessions are 90% corrective, 10% aesthetic. A good mastering EQ pass is subtle enough that most listeners couldn't identify what changed, but the track translates better on the next ten systems they play it on.
Compression controls the dynamic range, meaning the gap between the quietest and loudest moments. Light mastering compression (2:1 ratio, slow attack, slow release) glues elements together and adds punch without squashing transients. If you're pushing harder than that to make the track sound competitive, the mix probably needs work first. A limiter can't do what a compressor should have done.
Limiting sets the ceiling. A true peak limiter catches intersample peaks that clip during decoding and prevents distortion. It's also where loudness gets dialed in. For streaming, the target is -1 dBTP true peak and -14 LUFS integrated. Going louder doesn't get you louder playback. It gets you a squashed master that the platform turns down to the same level anyway.
After that: dithering if you're going from 32-bit float to 16-bit, and silence trimming to remove noise floor at head and tail.
The chain is short. The skill is knowing when to stop.
Streaming platforms have fixed loudness targets. Hit them.
Every major streaming platform runs loudness normalization. Your track gets turned up or down to match their target before it plays. If your master is louder than their target, it gets turned down. Dynamics preserved, but quieter in playback than you intended.
Platform targets in 2026:
Spotify: -14 LUFS integrated, -1 dBTP
Apple Music: -16 LUFS integrated, -1 dBTP
YouTube: -13 to -14 LUFS, -1 dBTP
TIDAL: -14 LUFS, -1 dBTP
Amazon Music: -14 LUFS, -2 dBTP
A master sitting at -14 LUFS and -1 dBTP behaves correctly on every platform. Push harder and you're not getting louder playback. You're losing headroom and dynamic range. The loudness war is over. The platforms won in 2016 and haven't moved since.
Note on Apple Music: its -16 LUFS target means a master you optimized for Spotify will be turned down 2 dB on Apple Music playback. Not a disaster, but worth testing the mix at both reference levels before you commit to a master that only sounds punchy on Spotify.
For club-focused tracks: the DJ's chain adds its own limiting and EQ. A track that sounds crushed on monitors will sound worse through a club system that adds another 6 dB of processing. Leave headroom in the master. This matters most for techno and drum and bass where the sub needs room to breathe and transient peaks define the rhythm.
The difference between mixing and mastering
These two get conflated constantly. They're different jobs with different scopes.
Mixing works on individual tracks. You're balancing levels, panning elements, applying per-track EQ and compression, building reverb sends, automating the vocal. The mix engineer has full visibility into the session and can pull up any track at any time.
Mastering works on the stereo output only. The mastering engineer receives a single bounced file, usually at 24-bit and the session's native sample rate. No stems, no individual tracks. Just the sum.
The reason they're separate: fresh ears on the final output catch things the mix engineer stopped hearing after 12 hours in the same session. A mastering pass often reveals tonal imbalances that became invisible through familiarity. The 2 kHz harshness that had been there since Tuesday. The low-mid buildup that accumulated across three different synth patches.
If you're doing both yourself, and most independent producers are, let the mix breathe for 24 hours before mastering. Export it, close the session, and don't open it again until the next day. You'll hear what you missed.
When AI mastering handles the job, and when it doesn't
AI mastering tools are the default for most independent producers now. LANDR, iZotope Ozone's AI assistant, and similar services analyze your track, pick a genre-matched processing chain, and return a master in under two minutes. For most tracks, this works well enough to release.
The cases where it holds:
Well-balanced mix with no major frequency problems
Electronic music with controlled dynamics
Streaming release with no vinyl or broadcast deliverable required
Budget under $15/month
The cases where it falls apart:
Mix with unresolved low-end buildup. The AI will master the mud louder, not remove it.
Track with inter-sample clipping in the mix bounced at 0 dBFS
Vinyl mastering, which has different loudness targets, different limiting approaches, and requires human judgment on sibilance and low-end mono compatibility
Broadcast deliverable with a specific loudness spec like EBU R128 or ATSC A/85
iZotope Ozone 12's AI assistant does well with in-DAW integration if you want control over the chain after the AI sets a starting point. Ozone Standard is $219 one-time. LANDR at $12.99/month is the cleaner option if you want a finished file with no DAW overhead.
The honest number: AI mastering produces a competent, streaming-ready result for roughly 80% of independent releases. The remaining 20%: dynamics-heavy recordings, complex acoustic material, releases with specific format requirements. Those still need a human making decisions about the chain.
Does AI-generated music need mastering?
Depends on the tool and what you're doing with the output.
Suno and Udio output audio that's already been processed to near streaming-ready levels. The generative model includes a final processing step in its pipeline. If you run the exported file through a loudness meter, you're typically landing at -13 to -15 LUFS integrated, which is within the range that works on every platform. True peaks are usually controlled.
Mubert's output for streaming is normalized to -14 LUFS by default, specifically built for Twitch and broadcast playback. You don't need to master it. You need to not over-process it. Running a fully normalized Mubert track through another limiter will add distortion, not presence. The tool already did the job.
Stable Audio exports at higher dynamic range. More headroom, more latitude for post-processing, but you'll need a mastering pass before uploading to a distribution service or playing on a PA system.
The check before you touch anything: measure the integrated LUFS and true peak of the export. If it's already at -14 LUFS and -1 dBTP, the track is streaming-ready. Adding another limiter stage creates distortion artifacts in the transients, not extra punch.
A free loudness check takes 30 seconds in any DAW with a built-in loudness meter, or you can drag the file into a web tool like MusicPulse's platform comparison tool to read integrated loudness against every platform target at once.
The workflow: from your DAW export to a streaming-ready master
For a track you've mixed yourself:
Remove any limiting or maximizer from the master bus before bouncing. Leave 3-6 dB of headroom.
Bounce at 24-bit, native session sample rate. Not 16-bit, not MP3.
Check for clipping. No signal should hit 0 dBFS in the mix bounce.
Import the bounce into a clean mastering session or a fresh DAW track. New gain staging from zero.
Build the chain: corrective EQ first, then compression, then creative EQ if needed, then limiting.
Target -14 LUFS integrated and -1 dBTP true peak. Use a loudness meter that shows both values simultaneously, not just peak levels.
Bounce the master at 24-bit 44.1 kHz for streaming distribution. Most distributors (DistroKid, Bandcamp, CD Baby) accept 24-bit WAV and handle the sample rate conversion for their specific delivery formats.
For AI-generated tracks: skip to step 6. Measure first. If you're already in range, export as-is and save yourself 20 minutes of processing a track that didn't need it.
The real test is the worst playback system you can find. The car speaker, the phone speaker, the laptop mono driver. If it still sounds like the track you intended, the master holds. Mastering that only sounds right on studio monitors didn't work.
Two cases where DIY mastering works: streaming-only electronic releases with controlled mixes, and lo-fi or bedroom pop where some roughness is part of the aesthetic. One case where it often doesn't: anything going to vinyl, where the lacquer cutting process and the playback cartridge physics require decisions that software doesn't make automatically.
If the mix is right, mastering is 30 minutes and a loudness meter. If the mix is wrong, mastering is three hours of fighting the same 200 Hz buildup. Fix the mix first.