Teach kids to read with ASR and computer vision. Reading Adventure from Osmo
playosmo.com#1 computer vision is used to detect what books the learner is reading to unlock digital content for each book (72+ books) and page (20+ Pages) as well and detect position of book relative to base/device/camera. Cv also tracks the wand which is used as a pointer to track the words the learner is currently pronouncing. The wand is also a pointer for certain games to practice what is being taught.
This is pretty neat! It's great to see relatively recent advancements in machine learning, put to use for child education.
A few questions for @Cherian:
1. I see the ASR usage, but where does computer vision come into play?
2. Are you training and/or fine tuning asr models to deal with the speech characteristics of children and new speakers?
3. Is the asr all cloud side, or do you have it running locally in some fashion?
Reading Adventure uses computer vision to help kids read through the books. The vision tech is used in combination with a “wand” to figure out the pages, words, and letters the kid is pointing to.
heyo,
2. The model is primarily focused around kids, with focus on repeated words and partial words
3. ASR is local to device. We are using a focused model per book
Thanks for all the answers and insight!
#2#3 Due to strict internal privacy policy, ASR is local to device and learner.
We plan to train and fine tune through continuous beta testing with families that opt in through our research groups but not through existing learners/subscribers.
Some other team members may comment to provide more info
Anyone in Bay Area welcome to come and check out demo.
Osmo is a favorite with Hacker News parents. Two years back, we started this labor of love to help them in this difficult journey of learning how to read.
Critique us. We’d love to hear from you.