SAN FRANCISCO--In-car voice recognition technology has come a long way, but there's much room for improvement. One problem voice-recognition technology faces is that is that there's only so much memory and processing power available in the dashboard at any given time, which has to be shared with navigation, media playback, traffic data, and other infotainment functions. Spansion, a company best known for providing flash memory products to OEMs, thinks it has the solution in its newly unveiled Spansion Acoustic Coprocessor, claimed to be the first of its kind.
The Spansion Acoustic Coprocessor is basically a system on a chip, combining a dedicated processor that handles the heavy lifting required for voice recognition and dedicated, non-volatile memory for containing the acoustic database of words and "fingerprints of sound" (to use the words of Spansion's CEO, John Kispert) that can be recognized and matched by the software. That acoustic database is supplied by Nuance, a name that we hear all the time in conversations about voice technology. The Spansion coprocessor's memory makes dedicated space available for larger and larger acoustic databases, enabling the voice technology to potentially recognize differences in gender, dialects, and multiple languages, which in turn enable more accurate matching of what that system hears with what it understands.
Spansion claims, and demonstrated at an unveiling today, that its Acoustic Coprocessor comes with the dual benefits of both halving the response time for the recognition of voice inputs and halving the load placed on your car's main processor. At today's demonstration, a side-by-side comparison between voice recognition with and without the coprocessor showed that Spansion was able to reduce the main CPU load by 51 percent and reduce the response time from 8.7 seconds to 3.4 seconds (a 62 percent drop).
Additionally, always-on technology means that the Spansion processor is able to spring to life and begin tackling the audio processing almost immediately -- a great claim to be able to make in the literally fast-moving world of cars.
Automotive hardware cycles are a bit longer than those of other tech verticals, so it may be some time before we see this technology in an actual car. However, Spansion claims that the Acoustic Coprocessor will be available for production and delivery to OEMs in early 2013.
Edit: Video demonstration added below.