Audible's New AI Narration Makes Audiobooks Sound More Human
Finally, an AI that can tell characters apart and keep you hooked for hours.
Audible’s technology team is using controllable text-to-speech to make audiobook narration more expressive and character voices more distinct and consistent. Unlike general-purpose TTS, which is designed for short utterances with minimal context, controllable TTS can follow natural-language descriptions like “speak in an enthusiastic, high-pitched voice.” Audible’s approach includes a data synthesis pipeline, a system that automatically profiles book characters and identifies their dialogue with 97% accuracy, and captions that act as stage directions for the model. In MUSHRA tests, this improved dialogue differentiation by 3.7 points, with statistically significant listening-experience improvements for eight out of ten voices across English and Spanish.
- Audible's new AI can follow natural-language directions to make speech more expressive and give each character a distinct voice.
- It identifies characters and their dialogue with 97% accuracy, so you can easily tell who's speaking.
- In tests, listeners rated the experience significantly better, with improvements across most voices in English and Spanish.
Why It Matters
This could make AI-narrated audiobooks more enjoyable and accessible, saving you money and time.