Skip to content

Media provenance & redistribution

Not-Happy-Jan docs

This document is the provenance manifest for every media asset Not-Happy-Jan redistributes through the media GitHub release (the bundles nhj setup-media downloads). It records what each asset is, how it was produced, and the terms under which it is redistributed. It is the publication gate for issue #52.

Summary

Media assets have mixed provenance — they are not all original recordings. Each group below is redistributed under the basis stated for it. The MIT code license does not cover media, and the media terms do not cover the code.

Most assets are either AI-generated (whose outputs the generator is licensed to redistribute) or synthesised procedurally in code. AI-generated audio is released into the public domain (CC0) — it generally is not copyrightable, and no third-party material is embedded.

Manifest

Asset group Files How produced Redistribution basis
Sound effects — handset, ring, tones, glitches, buttons audio/sfx/handset/*.wav, audio/sfx/{tones,glitches,buttons}/*.wav, audio/sfx/phone_ringing.wav, transfer.wav, failed_transfer.wav AI-generated with stabilityai/stable-audio-open-1.0 from text prompts (the generation tooling + exact prompts are retained in the project's private dev archive). Stability AI Community License grants the generator ownership of outputs. CC0 (public domain). No third-party samples.
Hold-tone clicks & censor beep audio/sfx/call_on_hold.wav, audio/sfx/call_off_hold.wav, audio/sfx/censor-beep.mp3 Synthesised procedurally in code (original DSP — the censor beep is a generated tone at the censor-beep frequency), phone-band filtered. CC0 (original work).
Hold music audio/music/*.m4a Generated with Suno under a paid plan that grants distribution rights. Redistribution permitted under Suno's paid-plan commercial terms.
Bundled voice bank (ships in the package) clips/{jan,bazza,karren,jan-whispering}/<state>/*.wavnhj/_bundled/clips/… AI-generated with Qwen3-TTS from the synthetic voice refs — bare-marker lines so a minimal install speaks in-character with no model download. No human voice. CC0 (AI-generated).
Character voice references voices/{jan,bazza,karren,jan-whispering}/ref.wav + ref.txt Synthetic — generated by a custom open-source TTS model (StyleTTS2/F5-TTS family), used as a zero-shot reference for Qwen3-TTS. No human voice is recorded or cloned. CC0 (AI-generated; no voice likeness).
Character voice lines audio/voice/**/*.wav Generated by NHJ's own TTS pipeline (synthetic character lines). CC0 (AI-generated).
Ambient beds audio/ambient/*.wav (beach, callcentre, outdoor, party, pub, special-forces-radio) Generated with Suno under the same paid plan as the hold music. Redistribution permitted under Suno's paid-plan commercial terms.

The dynamic LLM (separate download)

ocker-bogan-nano is not in the media bundle — it is downloaded from Hugging Face by nhj install-model. It was fine-tuned from an abliterated open base model on open data from Reddit and Australian forums. Its training data, base model, and license are documented on its model card.

User-provided voices

When you add your own character (voices/<name>/ref.wav + ref.txt), you are responsible for the rights to that audio. Only use voice references you created yourself or have explicit, documented consent and rights to use — voice likeness is legally and ethically sensitive. Not-Happy-Jan processes reference audio entirely on-device; your samples are never uploaded.

Bundle licensing

Each published bundle (voices.tar.gz, audio.tar.gz) includes a MEDIA-LICENSE.txt recording the per-group terms above, so the licensing travels with the assets independently of this repository.