Hey everyone,
While working on the upcoming v1.4 — which should bring quite a few new things, including VAD mode (continuous listening, no need to hold a key), faster transcription on CPU-only machines and on smaller GPUs, and better accuracy on long recordings — I came across a few bugs in v1.3 that needed fixing now rather than later.
So here is dictee 1.3.4, a maintenance release. What it changes for you:
- When you work on several transcriptions at once in the Transcribe window, each tab now behaves on its own — the audio player, the timeline, the rename field and the formatting choice stay attached to the tab you opened them in, even if you click around.
- Speaker separation no longer gets confused between transcriptions, and the formatting (with or without speaker labels) follows what you actually transcribed.
- The push-to-talk shortcut (F9 by default) no longer stays stuck after a crash or a forced reboot.
- Long audio files are now handled automatically: dictee splits them in pieces sized to your hardware, so you no longer need to worry about running out of memory on a smaller GPU.
Get it: Release v1.3.4 · rcspam/dictee · GitHub
You’ll find packages for Debian/Ubuntu, Fedora, Arch (AUR/PKGBUILD), a universal tarball, and the standalone .plasmoid widget. If you’d rather not pick, the one-liner installer does it all:
curl -fsSL https://raw.githubusercontent.com/rcspam/dictee/master/install.sh | bash
For those who don’t know dictee yet: it’s a fully-local voice dictation app for Linux, with native Plasma 6 integration (systray and widget). You can dictate in 25 languages, transcribe meeting recordings with speaker labels, and optionally pass everything through a local LLM to clean up grammar — all offline, nothing leaves your machine.
Don’t hesitate to share your experience — feedback, suggestions, anything weird on your setup. v1.4 will benefit from it.
Cheers,
rcspam

