Making Waves --- Sound Waves

By
iApplianceWeb
(08/31/01, 09:03:26 AM EDT)

The latest version of Wavemakers' ClearStream voice optimization software, v3, works in conjunction with automatic speech recognition (ASR) software so users of PDAs, cell phones, and Web tablets can use a more convenient voice interface, instead of the usual hunt-and-peck or handwriting recognition technology. ClearStream v3 is a single-channel technology that separates voice from noise in both close-talk (with a headset) and far-field (without a headset) situations.

Extensive testing demonstrates ClearStream's 'pure' speech signal improves the accuracy of leading commercial automatic speech recognition engines by up to 45%, reducing ASR errors by up to 100%.

"By increasing ClearStream's flexibility, precision, and power, we are advancing the ability of computers and devices to accurately understand and respond to voice commands," said Peter van der Gracht, Wavemakers' chair and CEO. "In high-noise environments, voice recognition accuracy hovers at a dismal 50%. With ClearStream, that percentage rises to 91.1%, elevating voice interfaces from a frustrating user experience to a positive one."

ClearStream v3's robust algorithms now easily recover drop-outs and handle sudden increases in noise found in real-world environments. ClearStream's enhanced voice detection capabilities translate into better detection of the start and finish of speech in noisy environments, especially when removing transient noises. Now, even background babble is removed by learning the target user's voice level and removing non-target speech significantly below that level. In addition, the integration of Wavemakers' new EchoBlock module with ClearStream enables barge-in for command-and-control environments and allows full duplex communication for hands-free voice applications.

Now available for embedded systems, ClearStream v3 works with Microsoft Windows, WinCE, Macintosh OS, and Linux operating systems. It also works with Intel StrongARM, x86, Hitachi SH-3/4, Power PC, and MIPS processor architectures.

Engineers working with ClearStream are also able to switch components on and off for particular architectures and, depending on the application, minimize processing and memory requirements on specific platforms. ClearStream's design ensures efficient memory usage and facilitates porting to embedded platforms. Wavemakers profiled ClearStream processor usage on an 800 MHz Intel Pentium III with 128 MB RAM, using a 60-second test file of continuous speech. At an 11 KHz sampling rate, processing required 22 MFLOPS and 1.7 seconds -- 36 times faster than real-time processing.

Link:
Wavemakers