AT&T might not be quite the monolithic Ma Bell of old but they do still have extensive research labs, and one project they’re working on promises to replace the monotonous narration of children’s ebooks with voices that are more distinct and entertaining.
AT&T held an AT&T Foundry event earlier today, and StorEbook was one of the projects that was shown off to press and other attendees. Engadget got a look at it, and based on their writeup I went looking for more info. I did some digging and unearthed more details as well as the promo video shows below.
The video stars Dr. Taniya Mishra, the researcher who is working on the project. As she explains in the video, the goal of StorEbook is to improve on the automated narration of a story so that it sounds more like a real person is reading the book aloud.
Rather than recite the words in a single voice like the current crop of TTS apps, StorEbook wants to make each character sound more like itself. For example, in the video the kids read the story of Goldilocks. In the case of Papa Bear, StorEbook would use a voice that was older and male.
The project is still under development, but other goals include making the entire process automatic as well as adding detection subroutines that would “make voices embody a character’s emotions so that a wolf sounds scary and a teacup sounds cute.” A related project is looking at creating a voice synthesis that could match a person’s voice by having them read a few hundred word script. This would enable a relative to read a story to a child without being present. It would also make it a lot easier to impersonate someone, but that is probably just a fun side effect.