Nativ background-audio exercise, 6 November 2026

These are original four-second mathematical tones, not voices, music licensed
from elsewhere, a generated translation or a Nativ source-separation result.
Use your normal comfortable playback level. Nothing plays automatically.

240/360 Hz tones = background bed; two soft chimes = sound effects.
750 Hz pulses = original-dialogue marker; 1000 Hz pulses = replacement marker.

Compare:
source-mix.wav: background + original marker.
new-only.wav: replacement only, with the background missing.
wrong-overlay.wav: background + BOTH markers, demonstrating the wrong input.
bed-plus-new.wav: background + replacement, with the original marker absent.

Separate ingredients are bed.wav, old-marker.wav and new-marker.wav.
All files are mono 16-bit PCM WAV, 16000 Hz, exactly 4 seconds.
No normalisation is applied between versions. They are not loudness-matched
listening trials or measured speech-quality examples.

To reproduce, download make_exercise.py and run with Python 3.10+:
python3 make_exercise.py my-new-exercise
The directory must not already exist. No external package or network is used.

Optional FFmpeg mixing exercise, from the directory containing the WAVs:
ffmpeg -n -i bed.wav -i new-marker.wav -filter_complex "amix=inputs=2:duration=first:normalize=0" -c:a pcm_s16le mixed-by-ffmpeg.wav
Both inputs already have the same length and start time. These levels were
chosen for this fixture; do not copy them as a production loudness preset.
The command mixes two existing tracks. It does not remove speech from a mix.

The exercise illustrates routing only. It does not test real dialogue,
separation artifacts, target-language timing, stereo positioning or Nativ
processing quality. Inspect the actual film and soundtrack when approving a dub.