Give AI a voice. Keep your own.
An experiment in expressive AI conversation, built for the person who wants to direct the performance rather than press generate and hope.
- 01Source material
- 02Editable script
- 03Directed voices
- 04Mixed conversation
The problem
AI-generated audio can be impressive right up to the moment you want to change something. A voice, an emphasis, a pause, an interruption. Starting again with another prompt gives the creator almost no control over the moment that matters.
The Wild Ducks part
Designed and built a document-to-conversation workflow, then extended it with multimodal speech generation and fine control over script, delivery and flow.
What was built
- A workflow from source documents and links to an editable conversation script.
- Speaker and delivery instructions for emotion, laughter, whispers, accents and pauses.
- Overlapping speech and interjections, with segment regeneration and separate audio tracks for editing.
Outcome
A working demonstration, published in 2024, showed the progression from controlled podcast generation to more expressive synthetic conversation.
Watch the demonstration (YouTube)
A creative proof of concept, demonstrated publicly in 2024. It is not a commercial service.
The interesting part of generative AI is not only what it can produce. It is how much room it gives a person to shape the result.