Storm cuts to sunrise

Make a 20 s music bed for a TikTok ad at 30 fps.
Output: one WAV, 48 kHz / 16-bit / stereo, exactly 960000 samples.
Python + numpy/scipy only. Synthesise everything (no samples, no downloads). Fixed random seed.
Timing: 80 BPM (1 beat = 22 frames). Cues: frame 0 = thunder hit; frame 240 = rain stops; last beat = soft ending that fades to exactly 0.
Sound: [chaos: thunder, wind gusts, rattling metal, noisy low drone] → [calm: bright morning pad, birds-like chirp synth, gentle 80 BPM kick], key G major.
Master: −14 LUFS integrated, true peak ≤ −1 dBTP.
You can't listen, so verify: measure loudness and peak with ffmpeg ebur128, check the exact length, confirm the energy change lands on each cue frame, check the notes fit the chords, render a spectrogram and inspect it. Fix and re-render until every check passes, then report the numbers.

Metrics

Integrated LUFS

-13 LUFS

True peak

0 dBTP

Length

20s · 960,000 samples @ 48000 Hz

Target master

-14 LUFS · peak ≤ -1 dBTP

Checks

Passed (runner)

Spectrogram

Generated code

Loading…