MADB reel: narration transcript
Source: madb_reel_acts1-2.mp4 (44.000 s, 1080x1920 H.264, mono AAC 44.1 kHz)
SHA-256: 33463a3c4e3b66605e30ff2ac13c401aa3114ee86370c4e2315bb9dc2be68c30

RESULT: NO INTELLIGIBLE SPEECH. THE SOUNDTRACK HAS NO SPOKEN NARRATION.

There are no spoken words to transcribe. No transcript has been written, and none
should be produced from the on-screen captions.

How this was confirmed (acoustic analysis of the extracted soundtrack, resampled to 16 kHz):
- 98.0% of total energy lies within +/-15 Hz of six fixed tones: 55, 110, 220, 261.6, 392
  and 880 Hz. These frequencies stay constant throughout. Speech produces moving formant
  bands and broadband consonant noise, and neither is present.
- Energy in the 300-3,400 Hz speech band, excluding those tones, is 0.17% of the total.
  Energy above 1 kHz is 0.03%. With the tones notched out, the residual is -38 dBFS,
  17 dB below the mix, and consists of encoder and envelope artifacts.
- Spectral flatness is around 1e-7 to 1e-3 in every 0.5 s window, which is purely tonal.
- The spectrogram shows only steady horizontal bands, a repeating pulse envelope, one
  short tone, one gap and one chord change. There are no syllable-rate patterns.
- webrtcvad (aggressiveness 3) labelled 80% of frames as "speech". It is known to fire
  on sustained low-frequency tones, so this was not used as evidence. The spectral
  measures above are what decide the result.

Timestamped audio log (non-speech; times +/-0.05 s):
00:00.000-00:18.550  Low synthesized pulse (55/110 Hz, repeating beat) with a soft hum (220 Hz); fades in over the first 0.5 s
00:18.550-00:19.200  Pulse cuts out; single sharp high tone (880 Hz) that decays quickly
00:19.200-00:19.950  Digital silence
00:19.950-00:37.450  Pulse and hum resume, same pattern as before
00:37.450-00:42.000  Sustained two-note chord (261.6 Hz + 392 Hz)
00:42.000-00:44.000  Chord fades out to silence at the end