This post was originally published on this site.
Those clowns in Silicon Valley are at it again. Despite being able to tell the difference between a bartender with a hot date and a bartender spending the evening alone ogling the ladies in the Sears catalog, AI still can’t replace everything humans do. One of those irreplaceable but soon-to-be obsolete people is Hank Azaria, the man of a thousand voices who has been making up the bulk of Springfield’s unruly mob since the late ’80s. Following his recent op-ed for the New York Times, Azaria put his vocal skills to the test in a new video for the Times, squaring off against an AI voice generator to see if this thing was a praise-worthy successor to the DJ Tron 3000 or a real useful invention like the Sarcasm Detector.
[embedded content]
Turns out, it’s more in the middle. Azaria shows off the range of Springfieldians he voices, including Snake, Chief Wiggum, and the Sea Captain. Still, when they put A.I. on the job, it offers a “vocal version of printed text,” albeit one that can’t pronounce “Azaria” correctly. Things get worse once Azaria asks the software to recreate the dulcet tones of Moe the Bartender to equally disappointing results—or celebratory results, depending on how quickly you want machines to replace human artists. It was “way off,” lacking the gravel and pronunciation that leads Moe to say “he’ah” instead of “here.” But, as Azaria points out, if they “were trying to sound like a robot, that would be a pretty good version.”
Azaria says he first became aware of A.I.’s threat to his livelihood as a voice actor when OpenAI released its latest SmarterChild clone with a voice eerily similar to Scarlett Johannson’s. However, he says that there is still a “humanness” thing that A.I. can’t do vocally yet, namely things tied to performance, like character motivation, emotion, and subtleties of physicality that are absent from the non-sentient experience. Technology firms are expected to spend more than a quarter trillion dollars on AI this year as companies continue to shoehorn this, frankly, unimpressive technology that serves no practical usage into all software. Thankfully, AI still has a long way to go toward becoming the labor-free everything app our tech overloads so badly want it to become.