0.6.0

Streaming dictation

• Transcribe while recording with local Whisper or OpenAI realtime (gpt-live-transcribe).
• --stream starts microphone recording; adding --record is optional. Desktop apps use the matching stream = true config setting.
• Keep recovery audio while streaming, cancel helper processes cleanly, and reject incomplete audio instead of pasting a truncated transcript.
• Apply dictionary corrections and polishing to the completed transcript, then paste once.

CLI

# Local Whisper transcription
hear --stream --engine whisper --model tiny.en

# OpenAI realtime transcription
hear --stream --engine gpt-transcribe --model gpt-live-transcribe

Add --no-polish to compare transcription latency alone. Whisper transcription runs locally; default polishing uses OpenAI. Use --polish-engine local for local polishing.

App configuration

In the app's existing config file, set:

[hear]
engine = "whisper"
model = "tiny.en"
stream = true

For OpenAI streaming, use engine = "gpt-transcribe" and model = "gpt-live-transcribe" instead. Restart the app after editing its config.

Streaming is opt-in. Existing file transcription and non-streaming recording remain available. Latency and accuracy depend on the model, hardware, and network; streaming is not guaranteed to be faster for short recordings.