x Abu Dhabi, UAESaturday 22 July 2017

Bad news gets through, even on garbled voice apps

Voice recognition software is like having a blunt, treasured friend around, to sort through the nonsense and the small talk and tell you how it is.

A lot of web and mobile phone companies now offer a cool service. They'll take your strings of voicemails and automatically transcribe them into text. There's an iPhone app that does a pretty good job of this, called Voxie. And it's also a service of Google Voice, the Google-brand telephone service that lets you use one number for pretty much everything.

The trouble is, I talk fast. And as anyone who's ever tried to have a complete and meaningful conversation with me knows, I also jump around a lot, from subject to subject, and I don't always telegraph a change of topic either. Sometimes I just start talking about something else, which I've been told is disconcerting to actual humans, so imagine how it baffles the voice recognition software that exists today.

For instance, I called a colleague of mine, Scott, last week, and left him a long voicemail, which was helpfully transcribed by Google Voice, and then e-mailed as text to his iPhone. The message was about a project I'm working on for a large television network, the script that's been delivered, and the chances that the project will be moving forward. Here's what Google Voice thought I said.

"Hey Scott, It's Rob. It's 11:30. It's got your message and Rob there at that. Excellent. That's really good. That's been a lot of the network he is the better. In general, I think you know script more prep we are just of. It's like you know, leave the handshake and the character to the fees. I think would be great. I guess that means. If you feel like if you just is. I'll forward it seems. Thanks! Bye!"

Utter nonsense, of course. So Scott dutifully collected the voice version of the message, listened to it, and called me back. The transcript of his voicemail was the following:

"Hi, Rob. I just want to and said basically you know what this is not the with regression for you to go and you should story lines and casting matter what number if you get to matter what arrangement you make it still gonna be. The project is not going forward, no matter what. You know I found out that you're not gonna it's a maybe we'll get a piece about that, but that's the only information I have. I was information. I'm gonna be sitting right network television few, so I'll talk to you later. Okay end up before like the correct."

The voice recognition software totally mangled up what I said, and delivered an almost incomprehensible text version of my voicemail. It then mangled up the response, and delivered a jumble of meaningless words to my e-mail inbox.

Well, almost meaningless. There was one sentence it got right: the project is not going forward, no matter what.

Which is true. And it came through clear as a bell, as if Google Voice, maybe the entire Google organisation, has heard those words before, knows them already by heart, has heard them mumbled and screamed and sobbed and stuttered and whispered like a deadly realisation, all over Los Angeles, and it has now adjusted its algorithms and voice-pattern-recognition computer code to recognise those words, in that order, no matter how they're spoken.

The project is not going forward, no matter what. Bad news, in Hollywood - and all over the world, let's be honest - has a way of making itself understood, despite static and slurred diction. Bad news, when we get it, is rarely a surprise - it's something we recognise, like a yearly bout of the flu or the sore throat we get when the weather turns wet.

It's hard to know for sure, of course, whether the Google computer brains are hyper-tuned for bad news, for the language of disappointment and rejection, or if - and this is probably the case - we just enunciate more clearly when we're handing out deflating information. There's a hemming preamble, some awkward stammering - no computer software in the world could effectively transcribe that - and then a clear statement of the negative: the project is not going forward, no matter what, followed by a lot of palavering nonsense meant to soften the blow, which the giant mainframe computers, arrayed somewhere in some desert location, won't deign to transcribe.

Voice recognition software is like having a blunt, treasured friend around, to sort through the nonsense and the small talk and tell you how it is. Finally, computer software that not only mimics human behaviour, but actually improves on it. It took my colleague several minutes to deliver the blow - the project is not going forward, no matter what - that my more honest and direct new best friend, Google Voice, did in a flash.


Rob Long is a writer and producer based in Hollywood