This is really cool. Voice cloning + translation in one pipeline is something a lot of content creators would pay for right now. Especially for YouTube dubbing where you want to keep the original personality of the speaker.
Are you handling the speech-to-text, translation, and voice synthesis as separate steps or is it more of an end-to-end model? Curious how you deal with things like pacing and intonation that don't always carry over between languages.
The proximity-based thing is cool, feels like actual walkie talkies but over the internet. The no-signup flow is nice too, a lot of voice apps lose people at the account creation step.
Curious what the latency is like in practice though. For push-to-talk to feel natural it needs to be pretty much instant. Are you using WebRTC for the voice transport?
2,684 requests on a single page load is wild. Would be interesting to see how much of that is just analytics and tracking scripts piling up on each other.
The redundant API call detection is something I wish more dev tools had built in. I've definitely shipped things where the same endpoint was getting hit multiple times on mount because of re-renders, and you don't notice until you actually look at the network tab.
The offline search with FTS5 is really nice. I have years of WhatsApp history and searching for anything in the app is painfully slow. Being able to just grep through everything locally would be a huge upgrade.
How far back does the backfill actually go? Does it pull your full history from the primary device or is there some limit?
Interesting approach! The IoT and smart TV tracking is something I haven't seen from other privacy-first analytics tools. Most alternatives focus purely on web.
Are you handling the speech-to-text, translation, and voice synthesis as separate steps or is it more of an end-to-end model? Curious how you deal with things like pacing and intonation that don't always carry over between languages.