AT&T Speaks Naturally
The capability to duplicate the human voice, ranging from famous personalities to generic vernacular, is becoming a reality. Today, AT&T Labs announced several new technologies aimed at making computer synthesized speech sound ordinary. AT&T's Natural Voices Text-to-Speech is divided into an "engine" that includes an archive of voices, and a solution to create customized voices for customers. The telecommunications giant has a long history of developing speech technologies; even still, the company faces stiff competition in order to achieve dominance in the field.
Whether it be resurrecting the voice of a famous celebrity to become a corporate spokesperson, or bringing back Gene Hart as the voice of the Philadelphia Flyers – a new market for sound technologies is opening up. In today's announcement, AT&T Chief Technology Officer David C. Nagel illustrated the company's intention to grab a huge portion of the emerging market. Nagel said, "We're going after the lion's share of what will be a billion-dollar market for text-to-speech systems in five years."
Holding more than 150 patents for speech technologies, AT&T hopes that devices ranging from cell phones to desktop applications will be willing to adopt its offerings. The company was quick to point out the social impact that speech technology has the potential to provide, claiming it can address the needs of individuals with disabilities. In a study performed by Benetech, a non profit organization, Natural Voices demonstrated a superior grade of performance out of 15 competing technologies.
Organizations such as Lucent, IBM, Lernout & Hauspie Speech Products, and Microsoft are also researching their own proprietary solutions. This added competition has forced companies to guard their intellectual property from infringement. Last June, AT&T filed suit against Microsoft for allegedly violating its voice compression patents.
When asked about potential ethical dilemmas resulting from machines, which have zero accountability, being able to perfectly match the intricacies of any given voice, AT&T brushed off the concerns. Spokesperson Kevin Compton told BetaNews, "The science of how this is done does not permit abuse." He continued on to say that a session with at least 20 hours of good tape would be required before a voice could be reproduced - making consent of the source a requisite. In cases were hours of tape already do exist for public figures such as Franklin Roosevelt, Compton claims, "The people who license the technology are certainly responsible for how it is used."