I released Humm[1] last week. Realtime speech-to-text with a focus on privacy. No backend, no telemetry - just a fast, nicely designed app wrapping local stt models.

[1] https://humm.so/