Music separation demonstration

Source: Core ’ngrato (1995), Singing Sergeants, United States Air Force Band.
Composition: Salvatore Cardillo. Lyrics: Riccardo Cordiferro.
Source and rights statements: https://commons.wikimedia.org/wiki/File:Core_'ngrato_-_Singing_Sergeants_-_United_States_Air_Force_Band.mp3
Original source cited there: https://www.music.af.mil/Multimedia/Music/Public-Domain-Music/
The source description identifies the composition, lyrics, performance and recording as public domain.

Adaptation: 12-second excerpt, 01:00–01:12, converted to stereo 44,100 Hz 16-bit PCM WAV, with 15 ms fades at both ends.

The vocals, drums, bass and other files are actual speech CLI HTDemucs INT8 results. instrumental.wav is made from drums+bass+other with the FFmpeg command shown in the guide. Outputs are illustrative samples, not a separation-quality benchmark.

Practice demonstrations (unreleased development CLI build; absent from release 0.0.29):
- slow-practice.wav: seconds 1–9 of sample.wav, speed 0.8, pitch unchanged, three repetitions; 30-second stereo 44,100 Hz PCM16 WAV.
- lower-key.wav: seconds 1–9 of sample.wav, speed unchanged, pitch down two semitones; 8-second stereo 44,100 Hz PCM16 WAV.
Both use the native offline time-pitch exporter, default 5 ms passage-edge fades and peak-only attenuation. These examples are not a perceptual-quality benchmark or proof of mobile-device performance.

Development-build commands:
speech practice sample.wav --start 1 --duration 8 --speed 0.8 --loops 3 --output slow-practice.wav --json
speech practice sample.wav --start 1 --duration 8 --semitones -2 --output lower-key.wav --json