"Speech recognition technologies offer a specific example of where we can start crafting specific policy and solutions for developing effective and equitable education technologies to support teachers and improve student outcomes," writes Russell Shilling. There are many ways speech recognition can fail: people speak differently as they age, people from different cultures may pronounce or use words differently, or people may have speech impediments. Failure to recognize some speech types may be depicted as a form of bias, and measures should be taken to ensure AI is less biased, argues Shilling. He focuses on a four part solution focusing on funding, quality, scrutiny and evaluation. I'm sympathetic, but it feels like an old-world solution to a new-world problem. Automated speech recognition (ASR) should be adaptive, generating individual personal models for each user, rather than being based on one model that is all things to all people.
Today: 3 Total: 20 [Direct link] [Share]
