There are 2.5 mil­lion people in the U.S. with severe speech dis­or­ders, and some 40 per­cent of them use speech devices to express them­selves. But those devices offer an extremely lim­ited selec­tion of com­put­er­ized voices.

We wouldn’t dream of fit­ting a little girl with the limb of a grown man — so, why then, a pros­thetic voice?” asked North­eastern Uni­ver­sity pro­fessor Rupal Patel in a talk at the TED­Womencon­fer­ence in San Fran­cisco last week.

Over the past six years, Patel has devel­oped a process for mor­phing voices in which she com­bines sam­ples of a patient’s speech — how­ever lim­ited it may be — with the voice of a donor who shares sim­ilar age, gender, size and location.

From the recip­ient, she extracts pitch, loud­ness and sibi­lance. As little as a single vowel may be enough. From the donor, she records a list of hun­dreds of utter­ances, which in the lab can be broken down to indi­vidual phonemes.


Read the article at All Things Digital →