Paul J Everett/Flickr (CC BY 2.0)

Machines learn to recognize baby talk

Watching a grown human talk to a baby can either be adorable or deeply unsettling depending on your perspective. Scientists studying baby talk have identified that people alter the rhythm and pitch of their speech, but now a new study in Current Biology suggests that the way we modify our voices for communication with infants goes even deeper than that—right down to the timbre of the voice itself. Timbre describes the texture of the voice—rough, smooth, nasally, breathy—and in a recent experiment, scientists were able to teach computer algorithms to identify mothers based on the unique vocal patterns of their baby talk. To make sure the programs were homing in only on timbre, the scientists recorded baby talk and normal speech from 10 different languages and used very short snippets of speech in the training phase. After the learning period, the machines could even distinguish between baby talk and regular speech with 70% accuracy, The Guardian reports.

Latest News