Getting near-100% speech recognition accuracy is fundamental to succeeding with voice technology in your warehouse. Proven solutions with more than just a few customers always use “speaker dependent” recognition software – meaning that the software is “trained” to recognize each specific worker’s different way of speaking, and a voice profile is created for every individual.
True, this training process does entail a little time and effort – but it pays itself back many times over in happy, productive workers whose performance is turbo-charged by great recognition rates.
In the rush to get new voice solutions to the market, many vendors have taken the shortcut of using a cheap, consumer-grade recognizer under the covers. You know, the kind of speech recognizer that often makes you repeat yourself on your credit card customer service line – even when you’re in a quiet room! This technology is not ready for prime time in a busy warehouse where you might have conveyors running, forklifts driving by, doors opening and closing, trucks starting up, and many other sources of variable (and loud) noise.
That’s why we are starting to see press releases from some of these vendors touting new abilities to “diagnose the cause of inaccuracies in recognition,” and new features enabling site management to “tune the recognizer for individual workers” (which sounds a lot like training). Who wants to pay a vendor to reinvent the wheel?
These developments are clear evidence that (A) consumer grade recognizers have problems in warehouse operation, and (B) speaker dependent technology still returns the best results.
|