Give that headline ☝️ to Amazon’s text-to-speech converter, and you’re in for a surprise.
The language processor sees right through your weak attempt at censorship and says that bad boy aloud.
We hilariously discovered this yesterday with a bot we’re testing that turns text from the Quartz app into speech.
Our bot scanned this (awesome) lead, converted it into speech using Amazon’s Polly service, and posted the audio in Slack.
“omg @johnkeefe,” bot editor Emily Withrow slacked me. “I listened to this 5 times because I thought surely this nice British man can’t be saying ‘Chipotle Mexican Grill ain’t nothing to fuck with.’ ”
He was. Put on your headphones:
Once I cleared the tears from my face, I did a little investigating and found that Polly doesn’t read other censored four-letter words, such as “s***.” Just, it seems, “f***.”
It’s possible to coax systems like Polly and Alexa into saying something else using speech synthesis markup language. For example, this SSML “bleeps” out the word:
<speak>Here are 5 stories to get you caught up: Chipotle Mexican Grill Ain’t Nuthin’ Ta <say-as interpret-as=”expletive”>F***</say-as> Wit: The fast-casual chain is banking on the Wu-Tang Clan to help it recover from its latest food safety nightmare.</speak>
So for anyone trying to protect their audience’s ears, that’s a substitution to consider.
Oh, and that’s not all we learned. Emily says that even before yesterday’s literal snafu, it’s been sport in her house to make Alexa say titles of Wu-Tang songs.