BY EVAN KOBLENTZ
Technology Editor, TMC Labs™

Dragon Takes Speech Recognition On The Road

Mobile professionals are paid primarily for their knowledge base. This means that when any of them -- or you, or I -- have a brainstorm, we need to note the idea as soon as possible, hopefully in textual form. But that's not always possible, because those ideas often come at a bar, over dinner, or while driving. One solution is transcription technology which, like most forms of speech recognition (and its cousin, text-to-speech), has only been perfected for niche applications.

But even if someone perfects dictation software, that's only half of the story for mobile users. A solution for the other half of the problem -- that of creating a mobile store-and-forward dictation device -- comes from Dragon Systems.

Dragon's NaturallySpeaking Mobile holds up to 40 minutes of voice or 80 minutes with an optional memory upgrade, in proprietary or .WAV files, organized in up to four folders of 99 files each. The device, according to Dragon literature (and presumably only after extensive voice training), can record up to 160 words per minute, alerts users to available memory through a backlit LCD, and connects to a PC with a serial port on the computer end and a USB connection on the recorder end. Three AAA batteries provide the mobile unit's power, and minimum system requirements include Windows 95, 98, or NT 4.0, a Pentium 133, and 32 MB RAM for Windows 9x or 48 MB RAM for NT 4.0.

Dragon engineers demonstrated NaturallySpeaking Mobile for myself and other TMC Labs engineers at a recent CTI trade show, and we're happy to report back to you that their bold feature and performance claims are true. After recording a few paragraphs of continuously spoken voice, they attached the device to a laptop PC, initiated a connection, viewed an organizer GUI, and downloaded the new file. Once downloaded, the engineers showed us how to save the file in a transcription format. Next, we opened the standard version of NaturallySpeaking, started the transcription engine, and watched as the software steadily began to write out our sample document.

As is normal for a well-trained dictation package, there were some errors, but we were impressed with the engine's knowledge of proper capitalization, proper nouns, etc. We were underwhelmed with the engine's speed, but the test PC was only a Pentium 233 MHz with limited available hard disk space, so obviously a PII or PIII system with more available storage will fare better. We were also impressed with the device itself, which has an ergonomic design, weighs very little, and uses only ten buttons.

For a street price of about $250 plus $60 for the memory upgrade, this device suite is a bargain. Like the current generation of PDAs and smart mobile telephones, these devices will eventually cost less, offer more features, perform better, and (for better or worse) probably end up running some version of Windows. Before all of that happens, however, I'd recommend this product to anyone who needs high-quality mobile speech recognition on a budget. Dragon developers clearly followed the keep-it-simple method when they designed NaturallySpeaking Mobile, and it pays off.

We plan to get this unit into our TMC laboratory for a full-blown evaluation, hopefully as part of a round-up among its inevitable competition. As is, it's a great product that solves a real problem. We look forward to hearing the stories of real-world users of this system, plus any suggestions for its improvement or notices about possible competition.

Evan Koblentz welcomes your comments at ekoblentz@tmcnet.com.