For this Friday fun a change, a kind of interactive game as it were, a kind of man-versus-machine test of wits (and hearing).
They're everywhere now. AI-generated voices that do everything from taking our over-the-phone pizza orders to scheduling our next doctor appointments.
So it was probably inevitable that websites would arise offering little quizzes consisting of audio snippets, one of a real person speaking, and the other of an AI app imitating that person's speech, and we are challenged to identify which is which.
Call it creepy fun. To hear both voices, one real, the other AI-generated, and try to figure out which is which.
But at the same time it can be somewhat disturbing to realize that the machines are moving closer and closer to impersonating us, to the point that we at times can't tell the difference.
Here's a website offering such a quiz to test our abilities to identify real-voice versus fake-voice. Call it up and take the quiz yourselves, and see how you do at identifying (or not) which is the real voice.
https://www.nbcnews.com/tech/can-tell-piece-audio-was-made-ai-rcna139769 BTW, I took the quiz and I got 7 out of 9 right, missing on Biden and Nikki Haley.
So did I use some kind of standard or tool to decide which voice was real and which was AI? No, not really. I just went with a "feeling" as to which of the two sounded more real, more like a real person.
Which begs the question, are there ways to tell, just from listening, whether a voice being heard is for real or is AI generated?
Well, googling
how to tell if a voice is AIproduces a number of sites offering guidance on this.
Here's one that mentions some tips one can use, what to look for (or rather, what to listen for) when trying to judge whether a voice being heard came from a real person or from an app:
https://marketingv20.com/how-to-detect-an-ai-generated-voice/It offers such guidance as:
"(AI is) robotic, monotonous, and lacking in personality or emotion."
" ... humans make mistakes while speaking, such as stuttering, mispronouncing words, and pausing to gather their thoughts."
"AI voices are typically flawless and don't make mistakes. If you hear a voice that sounds too perfect or smooth, it may be an AI generated voice."
"Listen for background noise. AI generated voices are typically recorded in a soundproof environment."
"AI voices often lack the natural inflection and emotion that come with human speech"
"They (AI voices) may also struggle with pronouncing certain words or phrases that are challenging to learn for the machine" Some other tips from other sites:
How good is the voice at sarcasm? (So far AI is challenged to voice a truly believable sarcastic tone of voice)
Are there unnatural pauses when the voice is replying to a question? A sign of an AI operating.
Does the voice slow down and then speed up, or vice-versa, even slightly, as it speaks (sign of a real voice)? Whereas the AI voice goes along at a more or less constant rate of speaking There are even apps available now (some free, some for a fee) that claim to use algorithms that can determine with a high degree of accuracy whether a given voice clip is from a person or from AI.
So it really is a brave new world out there, in terms of figuring out what's real and what isn't.
And it's more than just being amused by AI tricks.
"Deepfake audio" as it's known (as well as "Deepfake video") is already being used in criminal activities, to deceive people into paying ransoms for relatives they're convinced were kidnapped.
And the possibilities for their use in political "dirty tricks" campaigns are endless.
Other possibilities for misuse may be even more chilling. Here's an account of the use of a fake video/audio of President Zelensky of Ukraine from March 2022 (some say it was a product of Russian counterintelligence) supposedly calling upon his troops to surrender to the Russians. (The video is shown). It was caught and quickly taken off the web by Ukrainian authorities, but how many people might it have otherwise deceived by it, possibly even impacting world events?
https://www.npr.org/2022/03/16/1087062648/deepfake-video-zelenskyy-experts-war-manipulation-ukraine-russiaOr how about
video wills? Apparently they are not widely legally established, at least not yet, but if it happened that they were, how easy would it be to fake one, possibly entirely disrupting a person's final wishes?
The legal issues that these AI creations are already producing (celebrities are already bringing legal actions for what they see as abuses, for example) will need to be resolved, and hopefully soon.
So while they have their amusing aspects, the issues presented by the "deepfakes" will need to be worked through, until the "deepfakes" are settled into an acceptable place in cyberworld.
But in the meantime, they'll just keep getting better and better, until we really
can't tell the difference!
(BTW, if you'd like to have some more fun with this, human vs. AI, try the "AI Art Quiz" in the first link above, challenging one to decide whether a painting on screen was done by a fellow person or by an AI app).
Brave new world indeed!