OpenAI released the weights for their Whisper model, so you can run high quality speech-to-text, locally, at better than realtime if you have a decent GPU.