Got hit with a weird insight while training AI on my own voice recordings

So last week I uploaded 3 hours of me just talking to the computer to fine-tune a local voice model, and halfway through it started throwing back my own speech patterns at me. It nailed my filler words and pauses perfectly, but kept missing the sarcastic tone I use half the time. How do you train a model to pick up on context cues when humans barely understand them ourselves?

3 comments

3 Comments

perry.jesse4d ago

Yeah that thing about missing sarcasm hit me hard. I remember trying to get a smart speaker to understand when I was joking versus being serious and it just stared at me blankly. Reminds me of this time I tried teaching my buddy's parrot to say "you're hilarious" but he'd say it at completely wrong moments like during a sad movie. The whole idea of training something to read context feels like trying to teach a goldfish to drive a car you know? I don't even get sarcasm right half the time myself especially over text.

the_john4d ago

Three missed signals in one conversation? That's rough. I swear my phone's autocorrect makes me look like I'm having a stroke half the time. You think they'll ever make these things actually listen or is it just a lost cause?

shane1704d ago

Man, imagine if sarcasm detectors started calling us out for being sarcastic when we're actually being serious!