Correcting the Dragon

Looking over yesterday's dictated and Dragon-transcribed monologue, I'm encouraged by what appears to be a slow but steady improvement in its ability to convert my speech to the text you see here. Yesterday's dictation produced 590 words—although not all made it into the blog post—which required six corrections, making those claims of 99% accuracy increasingly believable. A couple were more misapplication than misunderstanding

This is where the computer models come in to into play, and although . . .

. . . another in accurate inaccurate forecast . . .

while others were very definitely in that misunderstanding category.

Even when you write you're right and the winter storm warning you issued . . .

. . . but make a mistake and the lame blame comes down like . . .

One was the result of my attempt to spell out the NOAA acronym one letter at a time, instead of just saying it outright. I didn't retry, but I've noticed the Dragon has a good grasp of acronyms—and jargon, generally—so it may well have recognized it, had I provided the opportunity.

. . . meteorologists at my local and oh AA NOAA weather headquarters . . .

Evidently, the letter N sounded like the word and in this case, but I guess that's pretty close, especially when the context isn't clear. The remaining correction was due to another one of those brief—and almost certainly unclear—sounds I so often make between words.

In a similar way, and an accurate weather forecast . . .

In fact, I probably was saying something that sounded very much like "anaccurate," and in cases like this it's best to avoid making the correction a permanent part of the speech-recognitions system's updated user files. Slurred words in the system's dictionary only add difficulty to the already challenging task of converting speech to text.

 

No comments:

Post a Comment