I’d updated Windows my Windows laptop to Windows 10 and Cortana – the Windows equivalent of Siri – kept suggesting I could ask her to do things for me.
Writing emails, short stories and articles seemed a nice option that would compensate for my lousy typing so I thought I’d see how Windows Speech Recognition compared to the alternatives. To do a totally fair trial I would need to test all the different systems at the same time so they all processed identical voice data.
With a bit of jiggling, I strapped two microphone headsets to my head, one going to my Windows 10 laptop and one to my work laptop where the VoiceNote2 plugin for Google Chrome was open.
I sourced a passage with 100 words exactly to I could work out % error rates. The passage is shown in italics below and the errors. The experimental design broke down afterwards because I wanted to test 4 systems simultaneously but only had 2 laptops to Dragon Naturally Speaking and Google Voice Typing were recorded separately but I attempted to keep the diction as close as possible between each session.
Do staff know how to create accessible Word documents, PowerPoints or videos? If not it is probably because nobody taught them how easy it is, what a big difference it makes to disabled learners and how it improves the experience of all learners. Fast forward to the day when a technology-aware disabled student comes to your institution having had excellent support with a previous learning provider. How long will it take to retrofit accessibility to your resources? Can you afford to buy in support to make an entire course accessible in a few days? Get it right from the start!
In the extracts below everything was correct except the text highlighted by red/italics
VoiceNote2 plugin for Chrome – no training. 99% accurate
Fast forward to the day when a technology away disabled student comes ….
Windows 10 Speech recognition – two sentence training. 93% accurate
His staff know how to create excess of one documents, cold winds or videos? If not it is probably because nobody taught them how easy it is, what a big difference it makes the disabled learners and how it improves the experience of all learners. Fast forward to of the day when a technology where disabled student comes to (missing ‘your’) institution having had excellent support with the previous learning provider …
GoogleDocs Voice Typing (no training) 98% accurate
NB. This was recorded separately so errors could represent different speech characteristics on the day.
If not it is probably because nobody taught them how easy it is, what a difference it makes to disabled learners and how (missing ‘it’) would improve the experience of all learners. Fast forward to the day when a technology away disabled student ..
Dragon Naturally Speaking (daily use and significant training input) 100% accurate.
Windows voice recognition is still, sadly, behind the rivals. It was fiddly to set up compared to VoiceNote2 and GoogleDocs Voice typing. It was also the least accurate. More broadly, getting Cortana to work with speech input required a lot of fiddling around, searching the inbuilt help then, when that didn’t solve the problems, searching the web then poking around in the Control Panel. It’s working now and will no doubt improve as it learns my voice but it had that familiar Windows clunkiness feel.
Nonetheless, there are millions of people who can benefit from speech recognition. From reducing RSI to allowing better focus on reasearch materials or even multitasking while writing there are excellent opportunities.
And even the worst of the bunch still saves you writing 93 words in every 100. That’s not a bad result…