Multi-Character Synthetic Narration Performs Well in Audiobook Study

A recent study conducted by Edison Research suggests that computer-generated voices are getting surprisingly good at storytelling, especially when books involve dialogue between multiple figures.

In a blind test of more than 1,000 fiction audiobook listeners, a majority of participants could not distinguish synthetic multi-voice narration from human performances. In fact, over 60 percent of listeners who heard the automated multi-voice sample believed they were hearing real actors. While human performers still received higher ratings for reading descriptive exposition, listeners expressed a clear preference for synthetic voices during multi-character dialogue scenes.

The findings indicate that listener hesitation toward automated audio quickly fades when presented with clear samples. Intent to listen to artificial narration more than doubled after participants heard the audio, and willingness to buy synthetic books matched human productions even after listeners were told the voices were software. These results suggest a future where independent authors can easily generate cast-style audiobooks, though it remains unclear whether listener interest will hold up across long, full-length audiobooks where artificial pitch and pacing might eventually tire the listener.