Speech Systems Debuts Innovative Client-Server System

It’s useful to repeat here the Pen-Based Computing mantra: ultimately, accurate speaker-independent speech recognition will be the keyboard killer, not the pen. Speech-based applications have obvious utility on the desktop, where people frequently enter prodigious amounts of information. However, mobile speech applications are even more appealing, especially since mobile workers are generally more functional when both hands are more or less free. And even when one or two hands are available, rarely are they trained for the delicate operation of keyboard input.

“Client-server-based systems may be the enabling technology for the emerging speech recognition market.”

The problem with speech recognition is that it tends to require significant processing power, considerably more than is currently available on mobile platforms. This problem is likely to continue in the near future given the trade-off between processing power and battery life that is characteristic of every system. Fortunately, there is another approach.

Enter Client-Server

The issue of local processing power is not a new one. In fact, much of the past decade has focused on delivering systems based on the model of reasonable local performance augmented with a high-performance back-end. Could this client-server model work with speech recognition? Boulder, Colorado-based Speech Systems, Inc. certainly believes so and recently announced a product enabling real-world distributed speech applications.

The product, dubbed SpeechNet, is a client-server version of the company’s Phonetic Engine speech recognition technology. SpeechNet is capable of operating over both wired and wireless networks, communicating with a remote Speech Server running on Microsoft Windows NT or Windows 95.

Here’s how the system works. SpeechNet Clients, which can be either hand-held computers or wearable speech-input devices, accept the speech input from the mobile worker through a microphone. The analog signal is converted to digital form, after which key speaker-independent features are extracted. This function requires the use of a digital signal processor supplied by DSP Solutions, Inc.

In order to make this information suitable for wireless transmission, the Speech Client compresses the acoustic data into a format known as a Phonetic Stream. It is this data stream, which requires a transfer rate of less than 2 kilobits per second, that is transmitted to the server. The Speech Server processes the input, converting the Phonetic Stream into text that is returned to the mobile application. At COMDEX, Speech Systems demonstrated this system running on a Fujitsu Stylistic RF pen computer, operating over a wireless local area network. Each Speech Server is designed to serve several Speech Clients on a single server.

Developing Speech-Enabled Applications

Using Speech Systems’ VoiceMatch System Development Kit or SpeechWizard development tool, programmers can create applications that fit into the SpeechNet client-server architecture. SpeechNet itself is compatible with all of the company’s Phonetic Engine platforms, including the PE500 system based on the ISA/EISA bus for desktop computers, and the PC Card-based PE1000 suitable for mobile computers. The system is designed in such a way as to allow the same program code to run in both the client-server and stand-alone environments, enabling you to leverage your investment across different physical structures.

Speech Systems, Inc.
2945 Center Green Court South
Boulder, CO 80301
Deborah Parsons
(303) 938-1110
(303) 9381874 (fax)
[email protected]