🟢 Quick Update on Picklewatch 🟢. I updated it to now RallyLab. I did this because initially i had an app to find open pickleball courts near me in Tampa. It worked, but I realized you really only use a court finder once. You check it, drive there, and that’s it. So I’ve been rebuilding it into RallyLab. The idea is simple: put your phone on the bench and let it listen to your match.
Instead of recording the match, the new app -RallyLab- just detects racket hits and the words you say like serves etc to detect who got the point. The audio gets processed 128 samples at a time and immediately overwritten, so there is no recording or audio file. 

The hit detector is working pretty well now. I tested it with simulated gym noise, talking, shoe squeaks, and another court playing nearby, and 9/11 conditions got an F1 of at least 0.93. I also added badminton, tennis, pickleball, squash, and table tennis, along with their different scoring rules. There are also 393 tests passing.

The biggest thing I’m working on now is automatic score tracking. I originally planned to have players tap who won each rally, but nobody is going to do that during a game. Since players already say the score out loud, I’m using that instead. Speech recognition by itself is only around 73% accurate, but because there are only a few legal scores that can follow each score, an HMM + Viterbi decoder can bring that up to around 96–100%. I said its almost ready but I am trying to make it realible for an athlete to trust.

Then I found a weird problem. The server’s score is always said first, so the microphone can know the score is 5-3 but can’t know which team is serving. That leaves two possibilities at exactly 50/50. I’m working on speaker ID now, but I’m using it as a probability instead of a hard rule so one mistake doesn’t ruin everything.
Next is getting the score system into the UI, finishing speaker ID, figuring out the iOS audio situation, and testing it in a real gym. So far all my testing has been simulated. If it can’t handle six courts playing at once, then I have a much bigger problem. Maybe in the future, once i get the grant from hackclub to buy a licence for playstore, the app might fucntion more efficiently than a web app.