Scottish voice recognition

[OC Richard]

This is pretty funny

https://www.youtube.com/watch?v=1EVnK-pNxaA

[O'Callaghan]

I've seen the one in the lift before. Of course, if you are an American you would call it an elevator to begin with.

I tried using voice recognition in my job, when I earlier tried self employment without ever having learnt to type. I found that Dragon Naturally speaking was fairly good, whereas frankly IBM ViaVoice is bloody rubbish for non-Americans. A friend of mine from Barrow-in-Furness (NW England) sent me examples of things the latter had typed when he spoke, and all you can say is they were hilarious.

Even though I found that Dragon Naturally speaking worked quite well, it did have one stupid problem for me. I am English and live in the US, so to get it to work properly I had to set it to its British accent. So what's the problem with that, you say? Just this - doing that ties it to British spelling, so 100% of the output had to be converted to American spelling by running it through the US spellcheck in my word processor, rendering the internal spellcheck entirely useless. And that's the good one. If I had the IBM product I couldn't think of anything better to do with it than using the disc as a frisbee!

I can't imagine how much worse it would be if I were Scottish, but there's no doubt of that. One of the Scots in the video gave us one small clue when he said he would have to set his iPhone to Doric, which unfortunately I don't think you can do. Historically, Doric was the language of lowland Scotland, and although very close to English, it's not quite English. Modern people in that region tend to speak a mixture of the two, or rather various mixtures of the two that vary in their proportions. Of course, most of them can speak a more standard form of English if they try, but even then their accent is not something that is understood by American mobile computing products.

Ironically, highlanders should have less trouble, because they speak either Gaelic or English, and not both at once, even though those two languages are very much further apart than Doric and English, or more likely because they are, although I'm not aware of any software that understands Gaelic.

[The Q]

One thing I've noticed in comments elsewhere is that it doesn't understand the Scottish accent . But there isn't a Scottish accent, the Gorbals area of Glasgow is significantly different from Kelvinside. As is Morningside, from the rest of Edinburgh. Accents change as you move around which is a nightmare for voice technology companies.

I'd agree dragon dictate was the better programme, I used to use it a lot often for updating dog pedigree files. There are a lot of made up names in pedigrees, as a breeder your put your affix on the front of the dogs name if you bought the dog in you put it on the end. All the normal registered affixes have long since been used hence the made up names. A common affix in rough collies is Corydon so I just taught that to dragon dictate and after that it recognised it every time.

I suppose you could teach the Gaelic to Dragon D but apart from taking a long time it would be very individual as the Gaelic varies tremendously between the islands.

Looking it up there are only 7 languages available for Dragon D so I would think that the Gaelic would be a long way down the list to be computerised.

[ForresterModern]

As a radiologist I dictated a hundred or more reports a day for more than 12 years using one voice recognition system that was a professional version (one of several) based on Dragon. That system was started with a sample reading of a couple paragraphs of standard verbage from which the program was supposed to get a basis of how each individual using the system said certain basic sounds, a basis which then over the next 100 hours of dictation matched in retrograde how words were pronounced with how they were subsequently spelled when corrected in the report (if it got it wrong the first time). So it learned as it went. To a point at least. We had several folks from the indian subcontinent (Indian and pakistani) as well as one Nepalese, all of whom spoke with a fairly heavy accent, heavy enough that often none of the rest of us 'mericans could understand them in conversation. But after the first week or so the program could pick them up with a 98-99% accuracy. I will try to check with my UK mates to see if they are using similar software and see how it works for them with the various UK accents.