AI Script-to-Voice – Automated Radio Broadcast Pipeline
A pipeline that turns a text prompt into a researched, AI-voiced radio segment and delivers it for broadcast on a schedule.
- Industries
The project
AI Script-to-Voice automates short radio segments, such as cruise-port weather forecasts and event bulletins, from prompt to air. A producer writes what the station should say in a plain text file. The pipeline sends it to OpenAI with live web search to write a script in the station's voice, turns the script into speech with Google Gemini and ElevenLabs, renders a timestamped MP3 and uploads it to a shared Google Drive folder that the playout system watches. It runs on a schedule or with one click.
The challenge
- Writing, recording and editing short recurring segments such as weather and news took staff time every day.
- Segments had to reflect what is happening now, so they could not be recorded far in advance.
- The station needed a process non-technical staff could control without code or dashboards.
What we built
- A prompt held in a plain text file that sets the cities, tone, length and sign-off, the only part a person edits.
- Script generation through OpenAI with web search switched on, so each script uses current information.
- Text-to-speech through Google Gemini and ElevenLabs, with automatic fallback to the other provider if one fails.
- MP3 rendering with segment names and timestamps, uploaded to a shared Google Drive folder for playout.
- Manual runs with one click, and scheduled runs through Windows Task Scheduler at any interval.
Key features
Plain-text prompts
Producers change what the station says by editing a text file.
Live research scripts
The AI looks up current conditions before writing the script in the station's voice.
Two voice providers
Gemini and ElevenLabs voice the script, and either can take over if the other stumbles.
Timestamped MP3s
Every render is named by segment and time, so no file is ever overwritten.
Drive delivery
Finished audio lands in a shared folder that the playout system picks up.
Scheduled or manual
Segments run every 30 minutes, a few times a day or on demand for breaking news.
