Spectral Mashup Engine (S42)

The mashup node is the showpiece of CutFlow's audio suite: feed it two tracks and it tries to make them agree on tempo, shift one's key, duck one under the other, and blend them into a single "mashed" mix - automatically. S42_SpectralMashupEngine is genuinely ambitious for a ComfyUI node, and it gets further than you'd expect. It also has sharp edges, because BPM detection and pitch-shifting are hard and the implementations here are pragmatic approximations.

Know what you're getting: this is a deterministic DSP chain - auto-BPM detection, time-stretch-to-tempo, pitch-shift, spectral ducking, and a mix balance. No model weights, no GPU, no magic. Which makes it predictable, and predictably imperfect.

How it works, step by step

1. BPM detection. Each track's tempo is estimated by autocorrelating its spectral flux - a real, standard method, though it's happiest with percussion-heavy music and will guess on ambient pads.
2. Master tempo. If target_bpm is 0, Track A's detected BPM wins; otherwise your value overrides it.
3. Sync. Both tracks are time-stretched (phase-vocoder style) so they run at the master BPM.
4. Pitch shift. shift_b_semitones transposes Track B by shifting its spectral bins (0 = in key, ±12 = an octave either way).
5. Spectral ducking. This is the interesting one - Track A's magnitude is used as a mask to carve Track B's spectrum down wherever A is loud, giving you that "the main track pulls out from under the vocal" sidechain feel, without any envelope followers.
6. Blend and normalize. mix_balance sets the A/B mix, and the output is peak-normalized.

Inputs

• track_a_main, track_b_layer - the two tracks. A is the anchor.
• target_bpm - 0 for auto (Track A's BPM), or a hard value.
• shift_b_semitones - -12 to +12. 0 = as-is.
• spectral_ducking - 0.65 default; how hard B ducks under A.
• mix_balance - 0.5 default; A vs B in the final blend.

Outputs

• mashed_audio - the result.
• detected_bpm_a / detected_bpm_b - the measured tempos, which are gold for debugging (and a sign of whether to trust the sync).

Installing it

Part of S42 CutFlow: ComfyUI Manager → search "S42 CutFlow", or git clone https://github.com/GeekyGhost/S42-CutFlow into ComfyUI/custom_nodes/, install requirements, restart. Needs torchaudio (bundled with the portable build).

Gotchas

The BPM detector is the weak link - two tracks with clean beats usually sync fine; anything with rubato, long intros, or sparse arrangement will produce a nonsense BPM, and both tracks get mangled to match it. The pitch shift is a bin-shift, so it's fine for keys but not studio-grade (expect some phasey smear). And there's no sample-rate conversion: Track B is assumed to match A's rate, and if not, the spectral math is off. Treat it as a sketchpad: set target_bpm yourself when the auto-detect looks wrong, keep mix_balance modest, and export the result to a real DAW for the final polish.