SuperAPI AI Directory

voice-and-audio

Design & creation

View repository

Updated 10-24 02:38

About this skill

Voice and audio

Always, as the last build step

Normalize Audio Level on every voice-track piece:

  • mode ITU-R BS.1770-4, target level -2.0 dBTP, target loudness -16 LKFS, set level Independent.
  • Check the result line ("normalized N pieces"); if it fails, give the creator these values to do by hand (select A1 → Normalize Audio Levels).

After normalizing, check for side effects

  • A silent clip (B-roll with no sound) gets boosted up to +30 dB → hiss. Set it back to 0 dB.
  • A B-roll clip with music/radio in it will fight the music bed: pull it down ~8 dB.

"My voice sounds muffled / underwater"

Measure before touching anything:

  1. Phase check: are left and right identical (true dual mono)? Compute L−R. If two different mics are summed (camera mic + lav), comb filtering causes exactly this sound: use one channel only.
  2. Spectrum: compare band levels of a 20 s talking section: below 300 Hz, around 1 kHz, above 3 kHz, above 6 kHz. A clean voice has its 3 kHz+ band roughly 10–15 dB below the total; we measured ~22 dB below with most energy under 300 Hz → the mic was covered (under a jacket) or badly placed.

Fix (most of it is EQ, which is not scriptable)

  • Scriptable: Voice Isolation 50 per talking clip (cleans room and noise; leave B-roll alone so its natural sound stays).
  • By hand (Inspector → Audio → EQ on one clip, then Paste Attributes with only Equalizer ticked, never Volume, so the normalize gains survive):
    • high-pass 80 Hz
    • −3 dB at 250 Hz, Q ~1.2 (boom)
    • +3 dB at 3.5 kHz, Q ~1 (presence)
    • +2 dB high shelf at 10 kHz (air)
  • Ask before driving the UI with computer-use; the creator may prefer to do it.

Prevention (tell the creator once)

Clip the lav on the outside of clothing, 15–20 cm below the chin; in stereo-mode receivers with one transmitter the voice is on one channel only.