DaimlerChrysler Scientists Develop Voice-Activated Computer That Masters Dialects
21 October 1999
DaimlerChrysler Scientists Develop Voice-Activated Computer That Masters Dialects
AUBURN HILLS, Mich. and STUTTGART, Germany, Oct. 21 --
DaimlerChrysler AG scientists have developed a computerized voice
recognition system that is capable of understanding human speech despite the
speaker's accent or tone of voice. The system can respond independently to
the speaker's instructions, without the need for human intervention. Compared
with conventional speech recognition systems, which only master individual
words and numbers, the system understands the content of entire sentences,
independently of the speaker and with no need for individual programming.
The voice recognition system is already in use in Mercedes C-Class, E-
Class, S-Class and CLK Coupe vehicles, and other applications are being
developed, such as telephone-access information centers, phone banking, and
unified messaging to manage all personal information systems with a single
software program.
The computer program, developed by the DaimlerChrysler's interdisciplinary
research group from Ulm, Germany, is programmed to understand the basic sounds
of speech, called phonemes. German, for instance, is made up of 40 such mini-
sounds, English about 50 phonemes. The software learns this readily
manageable repertoire of sounds, from which all words are assembled. The Ulm
scientists have catalogued speech data from the inhabitants of all regions of
Europe. The spoken words are separated into their individual sounds,
subjected to a training process and statistically evaluated to form the
comparative basis for the speech recognition system.
In use, the computer records the energy data and frequencies of the
individual sounds. It perceives the chain of phonemes as a series of
independent events; on this comparative basis, the computer investigates all
permissible combinations of phonemes and indicates the word path with the
highest probability.