A privacy-first Windows studio for turning any lawful script into polished QwenTTS narration mixed automatically over music.
RTAI Jingle is for everyone. It does not force a station name, brand, language, or script. Write your own copy, describe the voice in plain English, select a built-in royalty-cleared music bed or import audio you are licensed to use, and let the app engineer the result.
- RTAI Jingle installer — recommended for most people.
- Portable Windows build — unzip and run without installation.
Windows 10/11 64-bit is supported. The Qwen voice model is intentionally not bundled with the installer. On first run, the app explains the download, shows the destination folder, and waits for explicit consent. After the one-time model setup, voice generation works offline. Version 1.0 release binaries include the portable CPU runtime; it works without a dedicated GPU, although synthesis can take several minutes on slower computers.
- Narrates arbitrary user-written text with local
Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign. - Supports automatic language selection plus English, Turkish, French, German, Spanish, Italian, Portuguese, Russian, Japanese, Korean, and Chinese.
- Exposes an open-ended tone field and simple warmth, energy, and pace controls.
- Offers optional, blank-by-default pronunciation replacement for names and brands.
- Includes four attributed starter beds: Classical, Jazz, Pop, and Rock.
- Imports MP3, WAV, FLAC, OGG, and M4A music supplied by the user.
- Starts with music, ducks it automatically under the voice, restores it after speech, and finishes with a smooth fade-out.
- Denoises and compresses narration, sidechain-compresses the music, performs two-pass loudness normalization, and exports 48 kHz stereo MP3 at 320 kbps or 24-bit WAV.
- Runs without an account, telemetry, or a cloud narration service.
Copy: “Live from the city, independent voices and new music. You are listening
to Northline Radio.” Set tone to confident, contemporary FM imaging, choose
Pop, and use the Fast Hit preset.
Copy: “You are listening to The Research Desk. New evidence, clearly explained.”
Set tone to intimate, intelligent, close-mic podcast host, reduce Energy, choose
Jazz, and use the Balanced preset.
Copy: “The spring innovation festival begins Friday at the central auditorium.
Admission is open to everyone.” Select the required language, set tone to
warm, inclusive, clearly articulated, choose Classical, and use Cinematic.
Copy: “Freshly roasted every morning. Visit Harbor Coffee on Market Street.” Set
tone to friendly adult commercial voice, bright but natural, choose a licensed
imported bed, and adjust the intro/outro visually.
The following files are RadioTEDU demo IDs produced by the same general-purpose engine. They are examples only; RadioTEDU is not pre-filled anywhere in the app.
| Format | English | French |
|---|---|---|
| Classical | MP3 | MP3 |
| Jazz | MP3 | MP3 |
| Pop | MP3 | — |
| Rock | MP3 | — |
All demo music credits are recorded in examples/README.md and THIRD_PARTY_NOTICES.md.
- QwenTTS creates a high-resolution narration from the script and voice direction.
- FFmpeg denoises the voice with
anlmdn, filters it, and applies controlled compression. - The music begins alone and fades in.
sidechaincompressresponds to the actual narration waveform, lowering the bed during speech and releasing smoothly through pauses.- The music returns after the narration and fades out.
- A measured two-pass master targets −16 LUFS with a −2 dBTP ceiling before MP3 or WAV encoding.
Scripts and generated audio stay on the computer. Network access is used only when the user explicitly starts the model download from Hugging Face. The app does not collect analytics. Users must have rights to their scripts, voices, brands, and imported music, and must not use the software for impersonation, deception, harassment, or unlawful content. See Privacy and Security.
- User guide
- Academic methods, exact prompts, production matrix, and study design
- Build and packaging guide
- Contributing
- Third-party notices
- Changelog
git clone https://github.com/akgularda/rtai-jingle.git
cd rtai-jingle
py -3.12 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -e . -r requirements-dev.txt
python -m rtai_jingleFFmpeg and FFprobe must be on PATH for source runs. Release builds bundle both.
Use python -m rtai_jingle --self-test to verify resources without loading or
downloading a model.
Version 1.0 is designed around deterministic settings, explicit model consent, atomic output publication, cancellable background work, and automated unit plus real FFmpeg integration tests. See BUILDING.md for the exact test and release commands.
RTAI Jingle is an RTAI project. RadioTEDU contributed the original university radio imaging use case and the optional demo corpus. Bundled starter music is by Kevin MacLeod and is licensed under CC BY 4.0; exact attribution is embedded in the application and repository manifest.
Software source is released under the MIT License. Model weights, FFmpeg, music, and Python dependencies retain their own licenses.