technology-speech recognition

SPEECH RECOGNITION

Speech recognition along with VoiceXML and data integration via the Intranet/Internet
has enabled DAC Systems to bring to market a new breed of "speech enabled" IVR
systems.

DAC Systems has been working with speech recognition since 1989.

The current form of speech recognition is vastly improved over the early technology.
The advancements have been in several areas and they became first available in 1999.
Unlike the early recognizers which had to be acoustically trained for each word, word
recognition today is defined phonetically by typing works into a grammar list. The
grammar can be defined "on-the-fly" during each phone call. For example, DAC Systems
implemented an employee gift order entry application that brings up a list of gifts based
on the number of years of employment. The grammar is dynamic in that it configures
itself to understand the list of gifts specific to the employee and the gifts they can order.

The core recognition technology has improved due to higher performance
microprocessor and digital signal processing. SMC 4000 platforms with multiple
processors give the core recognition engine the processing power needed which was
impossible just a few years ago.

There are two primary types of speech recognition call flow - Directed Dialog and Natural
Language Understanding.

Directed dialog is essentially a question and answer session where the caller is asked
a direct question and they in-turn speak the answer. In this scenario, the speech
recognizer "picks up" a single answer per question. The following is a sample dialog
for a flight information system. "Would you like arrival or departure information". "I'd
like arrivals please." "Please say the name of the departure airport". "Chicago Midway."
"Please say the name of the arrival airport." "Orlando." "Around what time is the flight
due to arrive." "This morning." "One moment while we look up the arrival information.
There is a flight arriving at..."

Natural Language Understanding is characterized by the system understanding multiple
items in a single phrase. In the previous example, using the Natural Language
Approach the caller could say, "Is the flight from Chicago Midway to Orlando on time
this morning". At this time, most callers are not accustomed to such free form dialog
and it is difficult for the software engineer to predict and accommodate the multitude of
different ways a caller can pose their question. Over time, we expect our speech
enabled IVR systems to transition from the more popular and expected Directed
Dialog approach to Natural Language Understanding.

Speech recognition is available is many languages and is tailored to the regions it's
used in. For example there are several versions of English - North American, United
Kingdom, and Australian.