- Python 99.8%
- Makefile 0.2%
| tests | Improved exit, audio logic | |
| vosk_tui | Moved to package structure | |
| .gitignore | Added timestamp and export | |
| .python-version | WIP code | |
| LICENSE | Added system audio toggle | |
| Makefile | Moved to package structure | |
| preview.png | Added timestamp and export | |
| pyproject.toml | Moved to package structure | |
| README.md | Moved to package structure | |
| uv.lock | Added uv tool install | |
bladeacer's vosk tui
A TUI for speech recognition/transcription using Vosk.
Unofficial, not affiliated with VOSK.
Features
- Download your own model(s), select the one you wish to use
- Downsampling to 16 kHz for Vosk
- Timestamped Transcription Logs (max 100 characters per entry)
- Live preview of text about to be committed to logs
- Toggle between system audio and microphone. Defaults to microphone.
- Export current session as log file (appends)
Known issues
- System audio is currently hard-coded to regex for speaker.
- No configuration file
config.jsonandtranscript.logoutputs in CWD you are running it from
Setup
Install with uv
uv tool install git+https://codeberg.org/bladeacer/vosk-tui.git
vosk-tui
The executable will be built as vosk-tui.
Clone repo
Alternatively, you can clone this repository then cd into it.
git clone https://codeberg.org/bladeacer/vosk-tui.git
cd vosk-tui
Running in cloned repo
After cloning, you can chose any of the following for running the TUI.
- Use
uv syncanduv tool install .or callmake install
The Makefile flags are just helpers to perform the same action.
If you are using the script directly, remember to set file permissions if you are not on Windows
Running
uv syncand runuv run app.pyor callingmake runis deprecated
Download Models
Load your model directory directly e.g. vosk-model-en-us-0.22 instead of its
parent directory.
Find models to download at https://alphacephei.com/vosk/models/model-list.json.
Avoid models labelled with obsolete: "true"
Small models work better on limited hardware.
Usage
Ctrl Q to quit the app. Textual has the usual niceties.
Tested on Wayland/EndeavourOS, may or may not work on Windows.
Disclaimer
Created with the assistance of an LLM, human in the loop.
Vosk models are optimised for speech, might not work so well on stuff like music.
Contributions
Contributions of any kind are welcome. Do open an issue first before starting work on a pull request.
Credits
Audio Transcript recording taken from Harvard CS50.
License
Unlicense, see LICENSE.