By SongTools.ai
Automated mastering has improved significantly. But knowing when to trust it and when to override it is what separates good masters from great ones.
Automated mastering has been a contested topic in the audio engineering community since the first services launched around a decade ago. The early generation of automated masters was often criticised for being loud, compressed, and lacking the nuance of skilled human mastering. That criticism was largely fair.
The current generation is considerably more capable. AI mastering systems now use trained models to analyse each track's specific characteristics and apply processing that attempts to compensate for identified problems while meeting contemporary loudness and spectral targets. The results are consistently good enough for many applications — and understanding where the ceiling is helps you use these tools intelligently.
Consistency. AI mastering applies consistent processing logic to every track. For albums or EPs where tonal consistency across tracks matters, automated mastering will produce more predictably uniform results than a quick manual pass by an engineer who may have varying amounts of time for each track.
Speed and cost. A human mastering engineer costs money and takes time. For projects where budget is limited — demos, self-releases to streaming platforms, content that will be replaced within months — an AI-mastered version that's ready in seconds is genuinely useful.
Streaming loudness targets. Most AI mastering services are well-calibrated for streaming platform loudness normalisation. They understand the LUFS targets (-14 LUFS integrated for most platforms) and will deliver files that won't be turned down by Spotify, Apple Music, or YouTube's normalisation systems.
Identifying technical problems. Even when you don't use an AI master as your final output, running your mix through an AI mastering analysis will often surface technical problems — excessive true peak levels, very unbalanced frequency response, excessive stereo width in the low frequencies — that you can then address before sending to a human mastering engineer.
Musical judgment. The defining quality of great mastering is that it serves the music. An experienced mastering engineer listens to the track and asks what it needs — not what a statistical model suggests it should have. A piece of music that intentionally has a very dark, low-heavy frequency balance for artistic reasons should not be corrected toward a flat response. AI systems, without explicit instruction, will often push tracks toward average spectral targets regardless of artistic intent.
Context awareness. A mastering engineer knows whether they're mastering a vinyl release (which requires different low-frequency handling than digital), a broadcast master (which has specific loudness specifications that differ from streaming), or a music video soundtrack that will be compressed by a video platform's own processing. AI systems typically master for a single target context.
Handling problematic source material. When a mix has genuine problems — a resonant frequency that's causing pumping in limiting, a phase issue that's causing mono compatibility problems, or an imbalanced stereo field — a good mastering engineer diagnoses and addresses the specific cause. AI mastering tends to apply standard processing that works around the problem rather than solving it, often introducing its own artefacts in the process.
The final 5%. For commercially released music at the highest level, the difference between a technically competent master and an exceptional one often comes down to decisions that require not just good ears but context: how the final master sits in the context of the rest of the album, how it compares to reference material from the same genre, how it will translate to the widest variety of playback environments. These are judgment calls that current AI systems don't make at the same level as the best human engineers.
Use AI mastering for: streaming releases for independent artists with limited budgets where the goal is meeting technical minimums; demos being sent to labels, managers, or collaborators where the goal is demonstrating the music, not the mastering; tracks with near-final release deadlines where human mastering turnaround isn't feasible; and early-stage A/B comparisons where you want a quick, consistent treatment to evaluate different mix decisions.
Use human mastering for: major commercial releases, especially physical media (vinyl and CD require specialist treatment); projects where the sonic identity of the mastering is part of the artistic statement; tracks that have known technical problems in the mix that need engineering judgment; and situations where you need a mastering engineer's reference feedback as well as processing.
Before approving any master — AI or human — run a structured check:
Loudness. The integrated LUFS should be appropriate for your target platform. For Spotify and Apple Music, -14 LUFS integrated is the typical target. For tracks with heavy dynamic range in the music, slightly louder targets (around -12 LUFS) may be appropriate. True peak should stay below -1 dBTP.
Spectral balance. Play the master on multiple different systems: studio monitors, consumer earbuds, a phone speaker, a Bluetooth speaker. If significant problems only emerge on one of these systems, you have a balance issue the master hasn't resolved.
Mono compatibility. Collapse the stereo image to mono and listen. If important elements disappear, particularly in the bass or the vocal, you have a phase correlation problem that needs addressing before release.
Reference comparison. A/B the master against a commercially released reference track in the same genre at the same loudness. Don't focus on whether they sound the same — they shouldn't. Focus on whether the technical qualities (frequency balance, dynamics, stereo width) are in a similar range.
SongTools.ai's Mastering Check tool performs automated checks on integrated loudness, true peak, stereo width, low-frequency mono compatibility, and spectral balance. Running your master through this before release takes minutes and will catch most common technical problems before they're embedded in a released track.
One risk of AI mastering is the same risk that applies to any highly convenient tool: it lowers the barrier to releasing work that isn't ready. Because getting a master has become nearly frictionless, it's easy to move from a rough mix to a "finished" release in an hour, bypassing the critical listening phase where you'd otherwise identify problems that need attention.
The solution is discipline rather than technology change: make evaluating the AI master as structured as you would evaluate any other stage of the process. Let it sit overnight. Listen on different systems. Check it against references. Don't confuse the ease of producing a master with the quality of what you've produced.
The best use of AI mastering is as one part of a rigorous process, not as a reason to shortcut the process.
All the tools discussed in this article are available on SongTools.ai — free to explore.