YOUR STORY, AS A SONG
Generative AI & Audio
Global FMCG Brand
Media.Monks
Automation & AI integration
A WhatsApp campaign that turns a personal story into a song. A participant speaks about themselves and their life in a voice note; the system writes original lyrics from that story and composes a full song in their own language, then sends the finished track back over WhatsApp - all generated, no studio involved.
Project Overview
Festival campaigns are emotional, and nothing is more personal than a song about your own story. The brief was to let anyone become the subject of a bespoke song, in their own regional language, from nothing more than a spoken voice note - at the scale of a national WhatsApp activation, with no songwriter or studio in the loop.
The Challenge
- 01 Free-form spoken input - people ramble, mumble and go off-topic, and the story still has to become coherent lyrics
- 02 Regional-language quality - lyrics and vocals had to scan and sing naturally in the participant's language, not translated English
- 03 Brand and content safety - user-submitted stories run under a national brand, so every input needed moderation
- 04 Two-stage generation - lyrics then a full composed song, each slow, neither allowed to block the chat
The Solution: User Flow
- 1 User messages the campaign number, picks a language, accepts terms and gives their name
- 2 User records a voice note telling their own story
- 3 The voice note is transcribed and moderated for safety
- 4 An LLM turns the story into original song lyrics in the user's language
- 5 An AI music model composes and performs the song from those lyrics
- 6 The finished song is delivered back over WhatsApp as an audio track
Technical Architecture
Backend & Infrastructure
- n8n orchestration - lyrics and song as chained assets
- Node.js + TypeScript API, Postgres via Prisma
- QikChat WhatsApp Business API
- Per-asset generation status + async delivery
AI Services
- Whisper transcription of the spoken story
- LLM moderation of the transcript
- LLM lyric writing in the target language
- AI music generation for the final song
Key Results
One Voice Note
A Full Song
Lessons Learned
The magic is only as good as the input handling. A rambling, off-topic voice note still has to become a song someone is proud to share, so the transcription-to-lyrics step - cleaning, structuring and moderating a messy story - is where most of the quality actually lives.
Chaining slow generative steps behind a fast chat means treating lyrics and song as separate async assets, each with its own status, so the participant always knows something is happening while the machine works.