Files
transui/README.md
T
intelligence b161f81913
Build / build-amd64 (release) Successful in 44s
Build / build-arm64 (release) Successful in 56s
feat: compression
2026-08-14 14:10:41 +02:00

48 lines
1.5 KiB
Markdown

# Transui
A compact GTK3 desktop transcription tool for OpenAI-compatible and llama.cpp
`/v1/audio/transcriptions` endpoints.
## Dependencies
On Debian/Ubuntu install the following packages yourself before compiling:
```sh
sudo apt install cmake g++ pkg-config libgtk-3-dev libcurl4-openssl-dev \
libgstreamer1.0-dev gstreamer1.0-plugins-base gstreamer1.0-plugins-good libsecret-1-dev \
libjson-glib-dev wtype xdotool
```
`wtype` is used on Wayland and `xdotool` on X11. They are runtime tools, so
you only need the one matching your session. API keys are saved in the
freedesktop Secret Service (GNOME Keyring, KWallet Secret Service, and so on),
not in the settings file.
## Build
```sh
cmake -S . -B build
cmake --build build
sudo cmake --install build
```
## Commands
```sh
transui settings # configure endpoints, prompt, and destination
transui record # open the record window
transui # same as record
```
Bind `transui record` to a shortcut in your desktop environment. Transui does
not register a global shortcut itself. Press the same shortcut again to stop
and transcribe the current recording. Audio is recorded as 16 kHz mono MP3 at
32 kbps to keep uploads small.
The **Hosted first, local fallback** setting tries the Hosted endpoint, then
the Local endpoint when the first request fails. With it disabled, Transui uses
the Local endpoint only.
Both a normal response (`{"text":"..."}`) and llama.cpp event response
(`{"type":"transcript.text.done","text":"..."}`) are accepted.