goodroot/hyprwhspr v1.46.0

on GitHub

A very juicy release.

Thank you very much for all the new contributors. Your work and time is very appreciated.

New

Push-to-talk, hands-free

• Hold past push_to_talk_lock_seconds, release: recording latches on
• Press again to end it

Thanks @jonathancostin (#253)

New Parakeet variants

A faster .cpp variant and a fresh new Orukeet variant from our friend @Nathan-Roll1. Try it!

Parakeet.cpp (experimental)

• In-process Parakeet TDT 0.6B v3, multilingual
• Vulkan or CPU, no Python inference stack
• Pick it in hyprwhspr setup
• FASTER than the default. About as accurate.

Orukeet (Beta)

• Optional verified model for the ONNX backend
• Faster and more accurate than the default - perhaps the new default?

Thanks @Nathan-Roll1 (#258)

Setup

• One menu entry per model family
• Auto picks faster-whisper on CPU and NVIDIA
• Models compared by speed, accuracy, memory

Also

• debug_recordings: keep the last three raw takes, continuous mode too. Thanks @EnacheB (#265)
• clipboard_settle_delay for slow clipboards

Fixed

• Ducking heals apps left quiet by an ended stream. Thanks @JacobBanghart (#260)
• Polish layout types directly. Thanks @mmatczuk (#257)
• faster-whisper models found in any HF namespace. Thanks @neoscaler (#256)
• A quick stop or cancel sent mid-start is no longer lost
• Stop owns the microphone, even while start is settling
• Setup no longer carries old endpoints, headers or keys across a provider switch
• Re-running setup keeps your saved custom key

Under the hood

• Slow models load in the background; Ready arrives when they do
• One clean log line per event, no more interleaving
• Backends declare what they can do; no more name checks
• main.py split into focused modules, moved verbatim

Using push-to-talk binds? Switch key-up to release, then systemctl --user restart hyprwhspr.

Full Changelog: v1.45.2...v1.46.0