A developer posted a push-to-talk dictation tool to Hacker News this week, pitching it as voice input for Android apps and terminal workflows that only activates the microphone when you want it to.
The project page was down at time of writing, so the implementation details — which speech recognition backend it uses, whether processing happens on-device or through a cloud service, and how the Android and terminal integrations are wired together — are unconfirmed. The HN comments thread is the better source for those specifics until the page is back up.
The push-to-talk approach is a deliberate tradeoff against always-on listening. The microphone stays off until you trigger it, which eliminates false activations and keeps audio from leaking when you haven't chosen to speak. In terminal sessions that distinction is more than cosmetic — a stray activation that injects text mid-command can break things in ways that are annoying to undo.
How well it handles latency, which has historically made voice input awkward for technical work, is one of the practical questions the comments should answer.