Package Details: voxtype 0.7.3-2

Git Clone URL: https://aur.archlinux.org/voxtype.git (read-only, click to copy)
Package Base: voxtype
Description: Push-to-talk voice-to-text for Linux (optimized for Wayland, works on X11)
Upstream URL: https://voxtype.io
Licenses: MIT
Submitter: peteonrails
Maintainer: peteonrails
Last Packager: peteonrails
Votes: 12
Popularity: 1.81
First Submitted: 2025-11-28 17:08 (UTC)
Last Updated: 2026-05-19 19:28 (UTC)

Dependencies (30)

Required by (1)

Sources (2)

Latest Comments

« First ‹ Previous 1 2 3

graysky commented on 2026-03-13 23:20 (UTC)

Need to add git as a makedep:

  CMake Error at ggml/CMakeLists.txt:307 (find_program):
  Could not find GIT_EXE using the following names: git, git.exe

peteonrails commented on 2026-02-16 14:25 (UTC) (edited on 2026-02-16 14:25 (UTC) by peteonrails)

v0.5.6 released New features:

  • Moonshine transcription backend (experimental): 200x faster than Whisper on CPU for short recordings. Available in the Parakeet binaries, which now ship with both Parakeet and Moonshine engines. Enable with engine = "moonshine" in config and run voxtype setup model to download a Moonshine model.

  • Voice activity detection: Filters silence before transcription, prevents Whisper hallucinations on empty recordings. Enable with [vad] enabled = true.

  • Eager input processing: Starts transcribing audio while you're still recording. Reduces perceived latency for longer dictations.

  • Recording timeout now transcribes: Audio is no longer silently discarded when max_duration_secs is reached.

  • Append text after transcription: [output] append_text = " " adds a trailing space between dictations.

Bug fixes:

  • GPU isolation subprocess fixed (regression since v0.5.0)

  • Top-level --model flag now correctly forwarded to daemon

  • State file default path fixed

  • RPM package improvements (CPU-detecting wrapper, pipewire-alsa dependency)

Full release notes: https://github.com/peteonrails/voxtype/releases/tag/v0.5.6

peteonrails commented on 2026-01-28 15:05 (UTC) (edited on 2026-01-28 18:46 (UTC) by peteonrails)

New in 0.5.1:
- Parakeet engine: faster transcription, lower memory usage
- ROCm GPU acceleration for AMD GPUs (Parakeet)
- CUDA GPU acceleration for NVIDIA GPUs (Parakeet)
- Profile support: switch between configurations
- CLI backend mode for external whisper binaries

peteonrails commented on 2026-01-09 07:13 (UTC)

Should be fixed, autape

autape commented on 2026-01-06 08:45 (UTC)

==> Starting prepare()... error: the lock file /home/jc/.cache/yay/voxtype/src/voxtype-0.4.6/Cargo.lock needs to be updated but --locked was passed to prevent this If you want to try to generate the lock file without accessing the network, remove the --locked flag and use --offline instead.

peteonrails commented on 2025-12-18 05:14 (UTC) (edited on 2025-12-18 06:00 (UTC) by peteonrails)

v0.3.2 released

This release fixes the SIGILL crash on CPUs without AVX-512 (e.g., Zen 3, older Intel). The 0.3.0-2 fix didn't work because WHISPER_NO_AVX512=ON is ignored by the bundled whisper.cpp version. Now using compiler
flags (-mno-avx512f) to reliably disable AVX-512 instructions.

Other changes:
- On-demand model loading: optionally load/unload the Whisper model per transcription to reduce idle memory (~1GB → near zero)
- Paste output mode: copies to clipboard and simulates Ctrl+V for apps that don't work with virtual keyboards
- Build script portability improvements
- inotify integration fixes suspend-resume / screenlocker bugs

If you had issues with 0.3.0-2, this should work. The package now builds two binaries (AVX2 and AVX512) and the install script auto-detects your CPU.

peteonrails commented on 2025-11-28 22:33 (UTC)

Raphael - thank you for pointing this out. I've fixed the broken link to the source tarball

Dominiquini commented on 2025-11-28 21:43 (UTC)

Unable to download source. Error 404!

« First ‹ Previous 1 2 3