Magazine

At Last, A Phone That Takes Dictation


A wireless phone's main function is converting human speech into digital signals and back again. So why not harness handsets' speech-processing ability to solve a problem nearly everyone encounters when using cell phones for e-mail or other data applications? I'm referring to the difficulty of entering text on a dial pad, which speech recognition can remedy.

Manufacturers are finally getting serious about this solution. Use of spoken commands to dial calls and control other functions has been around for a while on some high-end handsets. But the Samsung SPH-p207 ($80 with a two-year contract from Cingular Wireless) is the first to incorporate speech-recognition technology for dictation of text messages. I found that, once I got the hang of the technique, speaking messages was much faster than tapping them out.

Based on earlier experiences with speech recognition, I was prepared to be disappointed. In the late 1990s, speech was promoted as an alternative to typing on PC keyboards. But the software never got good enough to make dictation easier than typing for most people -- although it has been a boon for people with disabilities or those who need their hands free while working. After Lernout & Hauspie, the leading speech-product company, imploded in a 2001 financial scandal, the industry retreated from consumer markets to focus on more lucrative niches, such as automated call centers.

PHONES HAVE ALWAYS BEEN A NATURAL for speech technology, but until recently handsets lacked the processing power and memory to make it work. The Samsung phone gets around that problem thanks to software from VoiceSignal Technologies that is economical in its use of resources.

One way it achieves that is through modest ambitions. Most dictation software aims for two goals: speaker-independence, meaning it will work without being trained to an individual's speech quirks; and continuity, meaning it understands normal speech, where words often run together. VoiceSignal took a different tack. The phone requires that the user spend about five minutes training the software, which involves reading a series words off the display. The speech recognition is then bound to that individual -- but given the personal nature of phones, that's a minor issue.

Also, the software only tries to understand one word at a time (known in the field as discrete speech). This means you must speak very deliberately, with a brief but distinct pause between words. It takes a little getting used to, but the payoff is accurate recognition.

To send a text message on the phone, you first press a button to put it into voice-command mode, then say "send text." When prompted, you either say the phone number or speak the name of someone in the phone's address book. If the addressee has both a cell-phone number and an e-mail address listed, you are asked which "number" you want to send the message to.

Once the addressing is done, you push a button on the side of the phone and hold it in as you dictate. When finished, you can move a cursor to any word the software got wrong. Press "0" and you can choose from a list of likely alternatives or tap in the right word from the dial pad. Then just press send.

The Samsung p207 is a clamshell phone that works worldwide on cellular networks running on the GSM standard. It's just 4 inches long, including a stubby antenna. It has all the features you would expect in a high-end handset, including a big, bright display, a camera, and a Web browser. But speech recognition is the feature that sets it apart from a couple dozen other attractive phones with similar functions.

Speech recognition failed to find a mass market in PCs because it wasn't as fast and accurate as using a keyboard and mouse. I think the same technology will succeed in handsets because the alternatives there are so much worse. The key is finding the right technology for the job.

For a collection of past columns and online-only reviews of technology products, click here

By Stephen H. Wildstrom


Tim Cook's Reboot
LIMITED-TIME OFFER SUBSCRIBE NOW
 
blog comments powered by Disqus