Juha Saarinen
Getting real with call centre voice technology
Attending tech conferences held overseas in 2020 usually means brainstraining under high cognitive load while watching video presentations late at night or in the small hours of the morning.
That’s Covid-19 for you, and I was reminded how right New Zealand got its response to the pandemic as I was able to go to the Auckland Amazon Web Services re:Invent mini-event in Auckland in person.
It wasn’t video-free, because we had to hook up with the Aussies who couldn’t travel across the ditch because of the quarantine, but that’s a minor quibble.
Re:Invent conferences are usually as massive as the constantly growing AWS is, and it’s easy to get lost in the cornucopia of new things.
Local heroes Sharesies and Xero talked cloud, and BNZ introduced its virtual AWS Connect call centre, featuring a customised voice, whina.
whina was created by humans training AWS’s Polly service that uses a neural network for machine learning. The whina sample I listened to sounds fairly realistic and is a welcome change from the usual North American robo-voices that you hear as you frantically tap buttons on your mobile to get to the right option.
And that is what BNZ says whina is for, the bank’s general manager of customer connection hubs Bridgette Dalzell and its programme manager Steve Aitcheson told me: to improve service and not to cut costs and call centre staff.
While whina uses ML, it isn’t an artificial intelligence bot that will start stringing sentences together based on what it learns from customers, Aitcheson said. One key feature for BNZ is that it’s easy to tweak what
whina says by typing new sentences into the Polly text-to-speak engine, so there’s no need to create new voice recordings.
Although cost savings weren’t the primary reason for BNZ to set up AWS Connect and whina, analyst firm Forrester churned through the numbers in a June report and reckons the mostly consumption-charged service with extensive automation saves 40 to 50 per cent annually, compared to regular call centre solutions.
What’s more, the AWS veep for business applications, Larry Augustin claimed that Connect can be set up in just minutes. This being the cloud, anyone with an Amazon login can try it out on the AWS site by going through some simple tutorials.
Connect doesn’t work with Safari and warns you to use the Chrome or Firefox browsers instead, and some things like the test Lambda functions are not available for New Zealand numbers. Other than that it was plain sailing. I soon had friends dialling into my call centre and hearing an overly effusive American voice telling them which option to press to speak to their friendly customer service agent. That was me by the way.
The nearest AWS region for NZ is in Sydney, but there wasn’t much delay. When you speak to people over Connect the audio quality is good and both callers can talk at the same time.
I was tempted to leave the IT helpdesk I created in Connect with a Lex chatbot which only advises people to reboot their devices to get them up and running, for all the cheapskates who use me for free technical support.
There’s the AWS consumption charges though. They may not seem much per unit of use, but AWS costs can be difficult to estimate, so the IT helpdesk had to die.
Nobody should have that much power anyway.
Configuring Polly, the AWS text-to-speech engine, is more complex but not much. Polly can be bilingual in some languages with lexicons added for custom pronunciation.
Some voices are generated through the neural network with 24,000 Hertz sampling rate, and sound pretty natural.
Te reo and suomenkieli (Finnish) are not offered by Polly, disappointingly enough.
It’s also possible to tackle Connect and Polly with coding which is again fairly straightforward. I’ve never even looked at these sorts of systems before and can believe that someone with experience would be able to set them up very quickly.
That’s the thing though: the low threshold to entry and AWS constantly adding new features — if something’s missing now, it’ll likely appear soon — and the ability to tie up with third-party customer relations management systems like Salesforce in the cloud make¯systemsA like Connect formidable offerings.
It’s the advantage that¯Aoperating at internet scale brings, and it’s difficult to see how even larger local providers and developers can compete against powerful, instantly available cloud offerings that¯Aare regularly refreshed.
There will be customisation work for nimble players, but it is a missed opportunity that we haven’t grown enough internet-scale companies.
Polly can be bilingual in some languages with lexicons added for custom pronunciation.