konradre/video-production-skills / skills/spot-audio-assembly
spot-audio-assembly
VerifiedAssembles the SOUND of a generated-video piece: voice-over takes (one voice and model per role), dubs for refused lines in the character's cloned voice, off-screen lines, sfx with the syllable map confirmed before the cut, music cues cut to markers with ducks, room tone, captions, the stem, and the read of the delivered audio. Use when a line must be voiced or re-voiced, a line was refused, sfx or music must be sourced or cut, captions must be built, a take's audio carries an artifact, or the delivered audio must be verified. Triggers — "generate the VO", "same voice as before", "dub the line", "the off-screen voice", "the sting", "the syllable map", "the music is too loud", "duck the music", "room tone", "captions", "is the vo on time", "musical artifact at the start". Not for placing lines on the timeline or the beat gate — use video-edit-edl. Not for the master render or the delivery encode — use video-finish-qc. Not for LUFS mastering of a finished mix — use mastering-audio.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 22.0 KB |
| references/CAPTIONS.md | markdown | 5.4 KB |
| references/MIX-AND-QC.md | markdown | 13.4 KB |
| references/SFX-AND-MUSIC.md | markdown | 13.2 KB |
| references/VO-PIPELINE.md | markdown | 6.6 KB |
| scripts/audio_head_scan.pyexecutable | script | 3.9 KB |
| scripts/beat_grid.pyexecutable | script | 30.1 KB |
| scripts/build_captions.pyexecutable | script | 8.8 KB |
| scripts/build_vo_stem.pyexecutable | script | 5.7 KB |
| scripts/gen_vo.pyexecutable | script | 7.0 KB |
| scripts/music_loop.py | script | 7.6 KB |
| scripts/music_take_read.py | script | 28.2 KB |
| scripts/qc_vo_placement.pyexecutable | script | 4.2 KB |
| scripts/roomtone_synth.pyexecutable | script | 3.4 KB |
| scripts/sfx_onsets.pyexecutable | script | 7.2 KB |
| scripts/sound_table.pyexecutable | script | 33.5 KB |
| scripts/vo_word_times.pyexecutable | script | 6.4 KB |
| scripts/voice_ab_reel.py | script | 3.9 KB |
scripts/sfx_onsets.py
View on GitHub#!/usr/bin/env python3
"""sfx_onsets.py — the syllable map of an SFX file BEFORE it is cut: onsets (a rise of ≥ --rise dB over the preceding
60 ms floor) listed by ordinal with their times and peaks, so a reel pick given in the operator's words — "the four before
the last one", "the end one, not the bitten-off start", "the second last one that starts at 0:43" — maps to sample-exact
cut points. --cut a:b writes the excerpt with 5 ms fades. Self-test first: a synthesized
four-click file must yield exactly four onsets, a whoosh its pass-by peak and a riser its end, or nothing else is printed.
--sync names where a sound's SYNC POINT sits inside the sound as it will play (after --src-in), so the sound is placed by
the VISUAL event it belongs to, never by a guessed offset: `transient` (a one-shot or an impact: its onset; --ordinal picks
which), `peak` (a whoosh: its pass-by, the loudest moment), `end` (a riser, a reverse cymbal, a swell: the hit it drives
into — the last moment within 1 dB of its loudest). With --event T (the visual event's timeline second) it prints the
`at` that puts the sync point on T, and the EDL's `sync` block. Pattern: mg-styles-15 `lib/audio/README.md` § 1.
sfx_onsets.py <file> [--rise 12] [--hop-ms 10] [--cut 0.83:2.10 --out audio/sfx/<name>.wav]
sfx_onsets.py <file> --sync transient|peak|end [--ordinal 1] [--src-in <s> --dur <s>] [--event <timeline s>]
"""
import argparse, hashlib, json, subprocess, sys
import numpy as np
SR = 48000
def pcm(f):
return np.frombuffer(subprocess.run(['ffmpeg', '-v', 'error', '-i', f, '-f', 'f32le', '-ac', '1', '-ar', str(SR), '-'], capture_output=True, check=True).stdout, np.float32)
def onsets(x, rise, hop_ms, min_db=-40.0, look_ms=100, refractory_ms=100):
h = int(hop_ms / 1000 * SR); n = len(x) // h
db = 20 * np.log10(np.sqrt(np.array([np.mean(x[i * h:(i + 1) * h] ** 2) for i in range(n)])) + 1e-9)
out = []; look = max(1, int(look_ms / hop_ms)); ref = max(1, int(refractory_ms / hop_ms)); last = -ref
for i in range(look, n):
floor = db[i - look:i].min()
if db[i] >= min_db and db[i] - floor >= rise and i - last > ref:
pk = float(db[i:i + look].max()); out.append((round(i * hop_ms / 1000, 3), round(pk, 1))); last = i
return out
def sync_point(x, kind, ordinal=1, rise=12.0, min_db=-40.0, hop_ms=5.0):
"""seconds into x of its sync point, or None. Levels from a 20 ms RMS on a 5 ms hop."""
if kind == 'transient': # 0.1 s of quiet ahead: a one-shot trimmed tight to its hit has no floor before it to rise from
pad = 0.1; on = onsets(np.concatenate([np.zeros(int(pad * SR), np.float32), np.asarray(x, np.float32)]), rise, hop_ms, min_db)
return round(on[ordinal - 1][0] - pad, 3) if 0 < ordinal <= len(on) else None
h, w = int(hop_ms / 1000 * SR), int(0.020 * SR); n = max(1, (len(x) - w) // h + 1)
db = 20 * np.log10(np.sqrt(np.array([np.mean(x[i * h:i * h + w] ** 2) for i in range(n)])) + 1e-9)
if db.max() < min_db: return None
if kind == 'peak': return round((int(np.argmax(db)) * h + w / 2) / SR, 3) # the loudest window's centre
return round((int(np.nonzero(db >= db.max() - 1.0)[0][-1]) * h + w) / SR, 3) # the END of the last window within 1 dB of the loudest
def selftest(rise, hop_ms, min_db):
rng = np.random.default_rng(3); test = 0.001 * rng.standard_normal(int(1.2 * SR)).astype(np.float32)
for k in range(4):
i0 = int((0.2 + 0.25 * k) * SR); burst = rng.standard_normal(int(0.05 * SR)) * np.exp(-np.arange(int(0.05 * SR)) / (0.01 * SR)); test[i0:i0 + len(burst)] += 0.5 * burst.astype(np.float32)
st = onsets(test, rise, hop_ms, min_db)
if len(st) != 4: return f'four clicks read as {len(st)} onsets {st}'
t = np.arange(int(1.0 * SR)) / SR; noise = rng.standard_normal(len(t))
whoosh = 0.4 * noise * np.sin(np.pi * t / 0.8) ** 4 * (t < 0.8) # passes by at 0.40 s
riser = 0.4 * noise * (t / 0.8) ** 3 * (t < 0.8) # drives into a hit at 0.80 s
tight = test[int(0.2 * SR):] # a one-shot trimmed tight to its first hit
got = {'transient': sync_point(test, 'transient', 2), 'tight': sync_point(tight, 'transient'), 'peak': sync_point(whoosh, 'peak'), 'end': sync_point(riser, 'end')}
want = {'transient': 0.45, 'tight': 0.0, 'peak': 0.40, 'end': 0.80}
bad = [f'{k} read {got[k]} (want {want[k]} ± 0.03)' for k in want if got[k] is None or abs(got[k] - want[k]) > 0.03]
return '; '.join(bad) or None
def main():
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument('file'); ap.add_argument('--rise', type=float, default=12); ap.add_argument('--hop-ms', type=float, default=10); ap.add_argument('--cut'); ap.add_argument('--out')
ap.add_argument('--min-db', type=float, default=-40, help='ignore rises that stay below this level (noise-floor flutter)')
ap.add_argument('--sync', choices=('transient', 'peak', 'end')); ap.add_argument('--ordinal', type=int, default=1)
ap.add_argument('--src-in', type=float, default=0.0); ap.add_argument('--dur', type=float); ap.add_argument('--event', type=float)
a = ap.parse_args()
bad = selftest(a.rise, a.hop_ms, a.min_db)
if bad: sys.exit(f'SELF-TEST FAILED: {bad}; the instrument is broken, nothing printed')
print('self-test ok: four synthesized clicks read as four onsets; a whoosh peaks at 0.40 s, a riser ends at 0.80 s')
x = pcm(a.file)
if a.sync:
i0 = int(a.src_in * SR); seg = x[i0:i0 + int(a.dur * SR)] if a.dur else x[i0:]
s = sync_point(seg, a.sync, a.ordinal, a.rise, a.min_db)
if s is None: sys.exit(f'{a.file}: no {a.sync} sync point found (too quiet, or fewer than {a.ordinal} onsets)')
print(f'{a.file}: {a.sync} sync point {s:.3f} s into the sound as played (src_in {a.src_in:g})')
if a.event is not None:
at = round(a.event - s, 3); print(f' at {at:.3f} puts it on the event at {a.event:.3f}')
print(' EDL: ' + json.dumps({'at': at, 'sync': {'kind': a.sync, 's': s, 'event': a.event}}))
print(' derived: ' + json.dumps({'from': [a.file], 'method': f'sfx_onsets.py --sync {a.sync} --ordinal {a.ordinal} --src-in {a.src_in:g}' + (f' --dur {a.dur:g}' if a.dur else ''),
'sha256': {a.file: hashlib.sha256(open(a.file, 'rb').read()).hexdigest()}}) + " (the plan entry's derived block: a swapped sound then FAILS the build until re-measured)")
return
on = onsets(x, a.rise, a.hop_ms, a.min_db)
print(f'{a.file}: {len(x) / SR:.3f} s, {len(on)} onset(s)')
for i, (t, d) in enumerate(on, 1): print(f' {i:02d} {t:7.3f} s {d:6.1f} dBFS')
if a.cut:
assert a.out, '--cut needs --out'; s, e = (float(v) for v in a.cut.split(':')); assert 0 <= s < e <= len(x) / SR + 1e-6, 'cut outside the file'
subprocess.run(['ffmpeg', '-v', 'error', '-y', '-ss', f'{s:.4f}', '-t', f'{e - s:.4f}', '-i', a.file, '-af', f'afade=t=in:st=0:d=0.005,afade=t=out:st={max(0, e - s - 0.005):.4f}:d=0.005', '-ar', str(SR), '-c:a', 'pcm_s24le', a.out], check=True)
print(f'wrote {a.out}: {e - s:.3f} s from {s:.3f}')
if __name__ == '__main__':
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 196k | Updated | Claude CodeCodex | Verified | Install |