v1.4 — Faster Transcription & Reliability

What's New

Faster Processing

• ~6x smaller audio uploads — audio is now downsampled to 16kHz mono before sending to Whisper (which uses 16kHz internally anyway)
• Whisper language hint — skips auto-detection by specifying English, reducing server-side processing time
• Short phrases skip GPT — transcriptions under 30 characters get find/replace only, saving a full API round trip

More Reliable

• Proper request timeouts — hard caps on total request time (not just idle time), preventing indefinite hangs on slow uploads or unresponsive servers
• Overall processing deadline — 150-second max for the entire pipeline, so it always completes
• Automatic retry — Whisper requests retry once on transient failures (timeouts, network drops, server errors)
• Cancel stuck processing — press fn again to cancel a hung transcription and start fresh

Better UX

• Instant Listening overlay — appears the moment you press fn, before the audio engine starts
• Step-specific status — shows Transcribing then Cleaning up instead of generic Processing