Private working material
In RTAI Jingle, scripts and generated audio remain on the producer's computer. The app does not require an account, telemetry or a cloud narration service.
RadioTEDU Artificial Intelligence
RTAI
AI that understands
the broadcast.
A technical workbench for live radio context, local voice generation and measured audio production, designed around a working student station.
01 Broadcast context
02 Local generation
03 Measured output
One name. Two production surfaces.
Models serve the programme.
They do not replace the broadcast.
RTAI connects live station context to the listener experience and gives producers a private desktop workflow for narration and imaging. Each surface has a narrow, inspectable role.
01 / Live context
A listener page that knows what the station is doing.
The RadioTEDU public platform accepts station handshakes, state snapshots, play events and cover artwork, then exposes station status to the listener interface.
Its published API contract is intentionally limited: it reports broadcast state and has no playout controls.
- Station status and now-playing context
- Play-event and cover-art ingestion
- English and French listener channels
02 / Local production
From a script to a broadcast-ready master.
RTAI Jingle runs Qwen3-TTS VoiceDesign locally, turns plain-language voice direction into narration, and mixes the result with a selected or licensed music bed.
- Automatic or explicit language selection
- Warmth, energy, pace and pronunciation controls
- MP3, WAV, FLAC, OGG and M4A music import
- 48 kHz stereo MP3 or 24-bit WAV export
Signal path
A visible production chain.
Every stage has one job. The output can be inspected, cancelled and regenerated without hiding the audio process behind a single button.
- 01
Write
User-owned script and voice direction.
- 02
Synthesise
Local QwenTTS narration.
- 03
Shape
Denoise, filter and controlled compression.
- 04
Mix
Waveform-aware sidechain ducking under speech.
- 05
Master
Two-pass loudness normalisation to −16 LUFS and −2 dBTP.
- 06
Export
48 kHz MP3 or 24-bit WAV.
Production boundary
Why Local Models?
Keep production close to the people making it.
Offline after setup
The voice model is downloaded only after explicit consent. Once installed, voice generation works offline.
Inspectable audio
FFmpeg stages, target loudness, source music and output formats are documented instead of hidden behind an unspecified service.
Honest constraints
Local does not mean instant. The model requires a one-time download, and synthesis can take several minutes on slower CPUs.
Public platform boundary
Status in.
Listener context out.
The public API describes a small status-ingestion surface. Audio playout remains outside this service.
RadioTEDU Public Platform · 1.0
POST /v1/radio/stations/{station_id}/handshake
POST /v1/radio/stations/{station_id}/snapshot
POST /v1/radio/stations/{station_id}/plays
PUT /v1/radio/stations/{station_id}/covers/{cover_id}
GET /v1/radio/stations/{station_id}/status
No playout controls
Open source and documented
Read the system, not a promise.
RTAI / RadioTEDU