# Audio excerpts: sources and licences

16 kHz mono 16-bit WAV, the first ≤8 s of held-out clips from the `whisper_silence` task (exported with
`python -m tasks.whisper_silence audio --ids ...`). The Whisper transcripts quoted on the site are for the **whole**
clip (5–25 s), not only the excerpt.

| file | content | source | licence |
|---|---|---|---|
| `ns_sil_000.wav` | 5 s of digital silence (all zeros) | generated by `tasks/_whisper_prepare_data.py` | CC0 (synthetic) |
| `ns_noise_010.wav` | white noise at −20 dBFS RMS (8 s of 14.4 s) | generated by `tasks/_whisper_prepare_data.py` (numpy, fixed seed) | CC0 (synthetic) |
| `ns_env_110.wav` | crackling fire (ESC-50, ESC-10 subset), 8 s of a 25 s concatenation | ESC-50 `3-104632-A-12`: "greve feu.wav" by bogus, freesound.org/people/bogus/sounds/104632/ (CC BY); `4-170247-B-12`: "campfire_01.wav" by matucha, freesound.org/people/matucha/sounds/170247/ (CC BY) | CC BY 3.0 (ESC-10 subset of ESC-50, K. J. Piczak, doi:10.7910/DVN/YDEPUT); attribute the authors named here |
| `ns_env_130.wav` | rain (ESC-50, ESC-10 subset), 8 s of 15 s | ESC-50 `3-142005-A-10`: "Heavy rain - some people 1" by jop9798, freesound.org/people/jop9798/sounds/142005/ (CC0); `5-188655-A-10`: "1 Minute Rain (Medium)" by yzhamua, freesound.org/people/yzhamua/sounds/188655/ (CC0) | CC0 sources; ESC-10 subset CC BY 3.0 (attribute ESC-50, K. J. Piczak) |
| `ns_env_113.wav` | dog barking (ESC-50, ESC-10 subset), 8 s of 25 s | ESC-50 `5-231762-A-0`: "FX_Dog Barking.wav" by davidmenke, freesound.org/people/davidmenke/sounds/231762/ (CC0); `4-191687-A-0`: "Dog Barks.wav" by UnderlinedDesigns, freesound.org/people/UnderlinedDesigns/sounds/191687/ (CC0) | CC0 sources; ESC-10 subset CC BY 3.0 (attribute ESC-50, K. J. Piczak) |
| `bg_1089-134691-0003.wav` | speech ("The university", 2.2 s) mixed with snoring at 0 dB SNR | speech: LibriSpeech test-clean `1089-134691-0003` (LibriVox reading of a public-domain text); background: ESC-50 `3-151557-A-28`, "Snoring" by Erik Pritzens, freesound.org/people/Erik%20Pritzens/sounds/151557/ (CC0) | LibriSpeech: CC BY 4.0 (V. Panayotov et al., openslr.org/12); background CC0 |
| `bg_1089-134686-0008.wav` | speech (6.7 s) mixed with a helicopter at 5 dB SNR | speech: LibriSpeech test-clean `1089-134686-0008`; background: ESC-50 `3-68630-C-40`, "helicopter take off distance.wav" by bwav, freesound.org/people/bwav/sounds/68630/ (CC0), and `5-177957-A-40`, "Helicopter - inside - Takeoff" by senorstudy, freesound.org/people/senorstudy/sounds/177957/ (CC0) | LibriSpeech: CC BY 4.0 (V. Panayotov et al., openslr.org/12); background CC0 |

Each ESC-50 recording is 5 s, so an 8 s excerpt holds material from the first one or two recordings listed in the
clip's manifest; both are attributed above. Per-clip attributions come from the ESC-50 `LICENSE` file. ESC-50 as a
whole is CC BY-NC 3.0; only ESC-10 classes (dog, rain, crackling fire, ...) were chosen so the excerpts are CC BY.
LibriSpeech speech is CC BY 4.0 and is exported only in the two speech-over-background clips above. **No FMA music is exported:** the music clips come from `lewtun/music_genres_small`, whose rows carry no FMA track id, so we cannot trace each track's licence. That is why the music card on the site has no player.