Computer Chips to Enhance Speech Recognition

Researchers at Carnegie Mellon University are using custom computer chips to improve speech recognition speed and lower its power consumption, as software has failed to overcome these problems.

Researcher Rob Rutenbar said, “Faster chip-based speech recognition will enable video players to search rapidly for Arnold Schwarzenegger saying ‘Hasta la vista, baby,’ in a movie. And lower power consumption will enable a cell phone to take dictated notes.”

Researchers on the university’s “in silico vox” project are currently working on two chip approaches, one using custom chips called ASICs (application-specific integrated circuits) and another using reconfigurable chips called FPGAs (field programmable gate arrays).

A videotaped demonstration was exhibited by Rutenbar. It showed the university’s technology using a low-end FPGA to recognize words in a limited 1,000-word vocabulary, and the system recognized several short sentences at about twice the speed it took for researchers to speak them. And the accuracy almost matched to that of Carnegie Mellon’s Sphinx speech recognition software.

According to Rutenbar, the researchers estimate their first-generation custom chip approach will be faster, to about twice the rate of regular speech for a 5,000-word vocabulary. They’re also working on a custom chip that is expected to work at 10 times the spoken rate. And later on researchers expect to include speed-up factors of 100 and 1,000.

The speech recognition chip converts an audio signal into combinations of noises that form any of about 50 different sounds, like “n,” in English. According to him, this is difficult as there are more than 1,000 sound possibilities.

The chip then compares those sounds to those used in actual words. Lastly, the chip looks for likely combinations of words–both pairs and threesomes—for better accuracy. The chip’s performance is reliant on high memory communications bandwidth and hence it can make comparisons quickly.

Source: News.com