It was Rod who started this by saying how great it is to replace slow typing with faster speaking and mentioning some kind of magical software package (appropriately named Dragon Naturally Speaking) which does transform your voice to written properly spelled English text with amazing accuracy. I was pretty sceptical, after seeing few years ago similar software from IBM with not-so-great accuracy, and even more recently the famous Vista speech wreckognition. Then Peter bought the software as well and had similarly good results as Rod. Now I got really interested. Both Peter and Rod and native speakers of English and I was very curious how will the software deal with my very own version of English pronunciation and fairly thick Slovak accent. So we did try it today.
Result: amazing. This thing works. I had to read for about five minutes of predefined sentences in training mode and during this period the software created “user profile” capturing personal deviations and details about the way I speak. After that I tried to read half page from a book (it was technical book on UML design) and Dragon got whole text correctly except one word. No misspellings, no typos. That’s much better result as if I would be trying to retype the paragraph by hand – and certainly much faster.
After that, just for fun I tried to say something not in English. I said something like
“Und jetzt werde ich ausprobieren wie es reagiert wenn man auf Deutsch spricht”
I am not 100% sure about the exact wording – but it was some German variation on “Now we will try out how it behaves whe I speak German” and here was the amazing result:
Let’s missed it a software sock and deny Gary tomato salsa and yet in a vendetta dish.
No, I did not make this up – it is copied from email word by word.
Of course, it is not a fair test, as the software was specifically set to US English. The company sells several language modules – including German, Italian and Dutch – which where not installed. I can imagine the result would be at least as good as with English – because English prononciation is probably most challenging for machine processing (why else would there be 3 separate modules – American English, Brittish English and Indian English). But it was fun anyway