I downloaded Nuance’s Dragon Dictation for the iPhone this morning on my way to work and gave it a try. It’s a free app that allows you to record short messages which are then processed by Nuance’s servers and returned as text that you can copy and paste into any other application. Dragon is pretty much the only game in town when it comes to PC and Mac dictation software, and those products routinely get great reviews from the press. However, my quick tests showed it to not be very accurate, or at least less accurate than Google’s excellent voice recognition in their app. Pondering this (and a few other things I noticed in the app) I’ve got a few ideas for them.
First, Nuance has said that although the app and service is free now, they may change that in the future. What they need to do is have a free app and a paid app. The free one would be the same as the one available now. The paid app would add one key feature: profiles. Profiles would allow the software to learn your voice just like the desktop applications. You would create an account with Nuance and every time you sent a recording to process the server would store it with your account. Over time the system would learn your particular speaking style and the accuracy would greatly increase.
This setup would have another benefit for Nuance: thousands of voice samples for them to study and use to improve their software. It would be even better if they could also get information on the corrections people made. The current app allows you to tap any word in the returned text and select a different word from a list of options (or edit it manually). I don’t know if the server returns the list of alternate words along with the text or if that is provided by a dictionary locally on the iPhone, but either way Nuance would be wise to have the app send them information on what words users changed. I would imagine having thousands of examples of words that the software got wrong, and the original voice recording that was processed, would be invaluable to Nuance. I didn’t see any indication of network access when I picked a new word from the list of alternatives so I don’t think they are currently sending that information back to them.
Oh, and the app definitely needs a clear button. I had to quit out and come back in to get a blank text box.