AI assistants work perfectly in the UK – unless you're from Cardiff, Glasgow, Liverpool, Birmingham, Belfast... To those of us born here, Britain is a wondrous cornucopia of accents and dialects. To visitors, like US-bred AI assistants, people outside of London may as well not speak English. Such is the conclusion reached by comparison website Uswitch, which found that smart speakers from the likes of Google and Amazon have difficulty …
Well, it also needs to understand the obligatory XKCD.
I can't even see it being useful unless you have a disability that necessitates it, in which case you've hopefully got better gear to begin with. Even without factoring in time spent correcting mistakes/mishearings, I can't see it being any faster than doing it manually the same way I can't imagine an audiobook ever being faster than reading.
"Frankly, do not bother with voice access things ever. Nokia had it working back in about 2008, Android/Google still cannot recognise a thing that is said to it in 2020."
I think that's because Google are using a sledge hammer while Nokia used a little toffee hammer. Nokia recognised a few command words and compared a name you spoke with what it found in the phone directory and did all the processing onboard the phone. Google appear to be using their full on "natural speech" cloudy parsing system.
On the other hand, Nokia phones often had "interesting" ideas about the names you tried to voice dial. One colleague had a name with a "gh" in it pronounced as "r" so to get my Nokia to voice dial him I either had to mispronounce the name or spell it phonetically in the phone book. On the gripping hand, Google can't find that name when spelled and pronounced properly either.
Oh, yeah, you're other statement. Voice dialling is good for in the car. Hands free and all that, often legally mandated in some jurisdictions. Luckily, in my case, the bluetooth car interface does voice dialling way better than "Hey Google" :-)
> people outside of London may as well not speak English
For many of us, people inside London do not speak English. For example the word button has two Ts in it. It is not pronounced "bu-un."
I once heard a Canadian comic perfectly describe his problem understanding Scots accents - we pronounce all the letters. The less Alexa, Siri, etc understand us the better imo.
That makes me think of the old advert for Tunes mentholyptus sweets:
A bloke with a cold goes to a railway ticket office and says
"Doo fird clad diggeds do Dottigab, Blead"
And the ticket seller goes "Whut"?
So the bloke goes and buys some Tunes, and has one, then goes back to the ticket window and says
"Two first class tickets to Nottingham, Please"
TUUUUNES!
These things never understand me either.
When using Google maps, my occasionally says "Sorry, I did not understand" - but I always to my best to turn off any voice anything, because I have never
known one to work.
The most annoying is the AVR used by banks. After asking you to say a number, it repeats it back to you and rarely gets as much as one digit in three correct.
Then it asks if it was correct, and you reply "NO" it assumes that you meant "YES".
In fairness, I cannot understand a lot of Americans on TV. I only discovered what "crosswalk" means after extreme harassment from Google's reCaptcha,
and today I was told what a popsicle is. If I am going to learn another language, Spanish is higher on my list than American - and so are Shona and
Greek.
But I am OK with Glaswegians, but my grandfather grew up in Edinburgh, and I once had a Glaswegian girlfriend.
Yes.
The longer answer is that even people need to learn or tune their ear to a new accent they've not previously encountered. So called AI speech recognition doesn't learn as it goes and both the providers and the users expect it to work without training it to the users voice.
I have great difficulties with some north east accents, went as far as once getting someone to write it down.
Another difficulty is mispronoucing vowls.
Seen one TV person every vowl was o as in got, so plont.
Some NI every vowl is i
My accent consists of replacing missing Rs on ends of some words.
Vowel softening and replacement is my blood pressure's No1 foe lately.
I thought the majority of it was just on my 7yo's telly, where-
"Look" has become "Lick" (fnarr).
"But" is now "Bert"
And "Put" is "Pert" (also fnarr)
But it's permeating normal telly too. *I* won't be able to understand anyone in a few years time, those illuminated bricks sat on the sideboard will have no chance. (Lar)
This post has been deleted by its author
But was funny.
I was ona conferene call with
Two from Glasgow
One from Liverpool
One from the black country
One from London
Two from Paris
One from Bacau (Romania)
One from Berlin.
About 10 minutes in we ended up in a fit of of hysterics when I pointed out the most used phrase was "can you repeat that, I didn't quite catch it"
Oddly it was the Romanians we all understood the best.
There's also the problem of people having multiple different devices, often from the same supplier in the room at the same time. Which one should respond? It's about time Google, Amazon et al allowed the users to set their own "wake" phrase like my 15 year old SatNav does.
"Hey Google, switch to BBC1". TV turns to BBCOne, smart speaker plays Radio One, phone asks if you want TV or radio, Alexa says "what's a google?"
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
