Voice (Mis)Recognition

Today I downloaded Microsoft’s Speech SDK v5.1. All 68 MegaBytes of it. I wanted to see how accurate it’s voice recognition could be. After all, talking into my script-writing program would surely be faster than my fairly hopeless typing.

I liked the fact that it worked for dictation and for commands. I thought it was cute that the demo dictation program that came with the SDK lets you switch from dictation mode to command mode by saying “computer”. Someone at Microsoft is a Star Trek fan I think…

Anyway, after many training sessions I thought I was getting good results, so I read out part of a draft scipt I am working on. Here’s what I said:

The attorney’s office is large and well furnished with dark leather and book spines shining in the light. The pompous attorney sits behind his desk with a piece of paper in his hand. He is staring expectantly at Isabella, who is seated opposite him. Isabella, in widow’s dress, stares back at him, a look of astonishment on her face.

Here’s what the Microsoft dictation software thinks I said.

If the attorney’s office’s lives and will face with the eleven and puts by shunning in the light. For all those attorneys 65 his desk the piece of paper is hand. His daring experiment with Isabella, who is seated opposite him. Isabella, Dean Witter’s stress, stays back again, a look for storage of the phones.

Interestingly, it got the name Isabella correct, which I wasn’t expecting. But otherwise not very accurate. If this kind of accuracy it typical, then it’s no wonder I cannot find many demo versions of competing products. I’d spend more time correcting than I do now.

Ah well, back to typing…