How Mixbot Masters Audio

Your original master file is probed for true container format, sample rate, bit depth, and channel count. We compute a SHA-256 fingerprint immediately upon arrival. Original uploads are marked immutable and sealed in private object storage.

The audio engine runs full BS.1770 multiscale integrated loudness, true-peak measurement, crest factor calculation, dynamic range evaluation, and sub-band frequency distribution across all musical sections (intro, verses, chorus, drops).

Mixbot formulates a deterministic MixPlan with explicit preservation budgets: maximum limiter gain reduction, stereo correlation thresholds, and harshness mitigation limits. If a frequency band is already balanced, the operator is explicitly bypassed.

The Python audio processing worker executes the plan against the source audio using floating-point 64-bit precision. Rendering is isolated: workers receive compute budgets and manifests, never customer auth tokens or billing states.

Before any master is presented to you, the rendered file is decoded back to raw samples and audited against 7 hard QC gates: true-peak ceiling (-1.0 dBTP), zero clipped samples, mono correlation compatibility, and transient retention.

In the workspace, you audition the original mix and Mixbot candidates with a common playhead, sub-millisecond crossfade, and automated browser audition gain staging. If you want adjustments, revisions are compiled non-destructively against the original audio.