On SpeechTEK's Second Day, the Future Is on Display
NEW YORK—The future is not that far off after all.
As it turns out, quite a few speech companies have been selling the future—promising technologies or system upgrades and improvements that do not exist in the hopes of selling present technologies with all their limitations, Bruce Balentine, chief scientist at Enterprise Integration Group, told attendees during the morning keynote session that opened the second day of the SpeechTEK conference Tuesday.
The time has come for them to stop selling the future and start focusing on the present, where speech technologies have become "good enough that they can deliver today," he said.
That's not to say that improvements aren't needed, especially in interactive voice response (IVR) systems, which Balentine says "still suck" after all these years.
A big reason that IVRs still evoke groans from consumers is that "they have been designed for the enterprise and not the consumer," he said. "And it was OK for companies to deploy bad systems because there would always be a lot of work toward improvements in the future."
Also holding IVRs back has been the design philosophy employed, according to Balentine. Companies, he said, have either tied their design strategy to an anthropomorphic philosophy that focuses on trying to build systems that sound and act like humans, or to a mechanopomorphic one, recognizing that it is a machine and limiting it to just that.
"These two worlds are merging into one with the ultimate goal of intelligent conversations," he said.
Along with that, there is a merging of directed dialogue, natural language, and artificial intelligence. The intelligence part of this is perhaps the most important and where the most work needs to take place, according to Balentine. "Speech does us no good if there is not some kind of intelligence that backs it up and makes it useful," he said.
For speech applications to be intelligent, they need to be self-correcting, self-monitoring, self-stabilizing, adaptive, and flexible, and should be able to learn on their own and learn even faster with some guidance, Balentine maintained. Intelligent devices, he added, have "complex problem-solving abilities."
And while in the past directed dialogue and natural language were at odds, today that is no longer the case. Speech systems can start with structured dialogue "that is natural and intuitive, and then after that, the open dialogue of natural language can happen."
And, it is in this future, which is perhaps no more than five years off, where conversational speech systems will grow and deepen and intelligent machines will become more common, Balentine concluded.
But for many others at the conference, five years is still a long way off, and many technologies that were once thought to only be possible in science fiction are now starting to gain widespread, real-world adoption.
Among them is smart home technologies. In one morning session, several panelists pointed to technologies that currently exist.
Mary Parks, principal experience designer at Honeywell, said her company already offers voice-enabled thermostats and home security systems and sees huge potential not just for consumer electronics for use in the home but also for commercial and industrial applications of the technology.
Jeanine Heck, senior director of product development at Comcast, said her company has already produced TV set-top boxes that respond to voice commands. Next, Comcast has its sights trained on devices that also offer voice output, as in