Mashups (mashups)

The mashups deck holds the station's branded re-edits: real source-song × title mashups (artist LOKLOK MASHUP) built by the content pipeline's mashup generator, plus user-recorded social mashups from mashup-engine.js. Files are MASHUP_<ts>_<n>.mp3 in media/mashups, Break-tagged and R128-mastered.

Two mashup generators

Generator	Source	Output
Pipeline mashup block	Real songs from the music decks	MASHUP_<ts>_<n>.mp3, real titles, artist LOKLOK MASHUP
mashup-engine.js	User-recorded clips + bed + sfx	SOC_MASH_<ts>.mp3, artist LOKLOK MASHUP

The tempo-lock equation

Pipeline mashups blend ≥ 3 songs whose BPMs must be compatible. Each track's BPM is measured (detectBpmAndOnsets), a target is chosen as the median of the sorted BPMs, and every track is resampled by:

$$r = \frac{tgt}{bpm_i}$$

The tight band requires 0.96 ≤ r ≤ 1.04 (a ±4 % lock — beyond that, voices sound wrong). If not enough tracks land in band, a relaxed band 0.92 ≤ r ≤ 1.08 is allowed; otherwise the generator picks the 3 tracks closest to target. The mashup is assembled on an overlapping timeline, mixed, limited, and R128-mastered to −14 LUFS / −1.5 dBTP.

How the scheduler slots them

Mashups get a guaranteed cadence slot — without it they were starved to 0–2 per rotation because the VAE pro-suggestion and anti-fatigue overrode the raw MASHUP_CHANCE:

$$N_{MASHUP} = \max\left(6,\ \left\lfloor \frac{\text{targetSize}}{\max(2, \text{mashupCap})} \right\rfloor\right)$$

≈ every 8th slot → ~12–14 mashups per 120-track rotation, skipped during deep night (02–05) and when the previous slot was already a mashup. A stinger frequently leads the mashup in the same slot (the stinger bridge).

• Music cooldown (3 h) applies — mashups are treated as music, so the song-key dedup does not apply to them (they are legitimately recurring).
• liq_cross_duration: none — mashups keep the full 2 s crossfade.

User mashups (mashup-engine.js)

For SOC_MASH_* builds: up to MAX_CLIPS = 4 recordings (≤ 6 s each, min 1.0 s), overlapping by OVERLAP = 0.15 s, over a bed at BED_VOL = 0.30, each clip gets one random effect (dropout / pitch / echo / flanger / phaser / vibrato / reverse / lo-fi), junctions get stingers, tail padded TAIL_PAD = 1.5 s. Same overlapping-adelay → amix → two-pass R128 finish.

Verify on air

1. media/mashups holds MASHUP_*.mp3 with real source-song × titles.
2. m3u entries for mashups carry no liq_cross_duration override (full 2 s crossfade), annotate:type=mashups.
3. Rotation has ~12–14 mashups per 120 tracks, none back-to-back.
4. ffprobe: Break-tagged, R128 ≈ −14 LUFS; BPM-resampled tracks audibly in-tempo with each other.