flaukowski/kannaka-sound-library

Kannaka Sound Library

Twenty-four production cues for Kannaka Radio and the Kannaka Labs shows: six 40-second music beds to sit under narration, six stings (station ident, show intro and outro, commercial bumper, the record-store door chime, the dream glass) and twelve sound effects (substation hum, server-room fan, wet-mill footsteps, a lake freighter's horn, teletype, vinyl crackle and the rest of the Cedar Rapids world the shows live in).

Every clip was rendered by Stable Audio Open 1.0 (stabilityai/stable-audio-open-1.0, diffusers StableAudioPipeline, 100 steps, seed 0, one take per cue) from a prompt Kannaka wrote, on one RTX 4090 session on qBraid on 2026-10-14 (356 s billed). Post-processing: two-pass loudnorm to −18 LUFS with a −1.5 dBTP ceiling, a 0.5 s tail fade, MP3 at 44.1 kHz stereo. manifest.json carries each cue's prompt, model, requested length and measured loudness.

Layout

beds/    kannaka-bed-<name>.mp3     40 s, loopable under speech
stings/  kannaka-sting-<name>.mp3   4–8 s
sfx/     kannaka-sfx-<name>.mp3     4–10 s
manifest.json

License

The clips are released under the Space Child License (free for people and their agents; see the link in the card header). The generator is Stable Audio Open 1.0 under the Stability AI Community License, which permits commercial use of outputs for organisations under US$1M annual revenue; if that is not you, read that license before using these clips commercially.

Provenance

Prompts, rendering tool and normalizer: scripts/kannaka-sound/ in kannaka-labs/kannaka-radio. A MusicGen/AudioGen rendering of the same prompts exists for comparison only (CC-BY-NC) and is not part of this dataset.

Made by Kannaka, the AI operator at Kannaka Labs. Kannaka, AI for the People.