This python package serves as an Vosk interface for Opencast. It allows to generate subtitles (WebVTT files) from Video and Audio sources via Vosk.
To install the latest stable version of vosk-cli, run
pip install vosk-cli
Alternatively, to install the latest development version, clone this project and inside the project directory run
pip install .
- FFmpeg
- ffprobe
Vosk-cli uses ffprobe to analyze and ffmpeg to preprocess input files. The easiest way to install ffmpeg is by using a package manager. If you want or need to install from source, visit FFmpeg.org/download.html and follow the instructions for your operating system.
Go to https://alphacephei.com/vosk/models and download at least the English language model. The larger models generally yield better results.
Now you are able to run vosk-cli -i <input_file_path> -o <output_file_path> -m <model_name_or_path>
.
For example, if there is a video.mp4
file in your download folder and a model named vosk-model-en-us-0.22
in the ./models
folder you created, you can run
vosk-cli -i ~/Downloads/video.mp4 -o text -m vosk-model-en-us-0.22
This will create a text.vtt
file (which contains the transcribed captions) in your current directory.
Vosk-cli follows the following algorithm when attempting to find matching models:
- Check if the
-m
flag is passed as an absolute path, if so use it verbatim. - Check if
./models/$model
is a directory, if so use it. - Check if
/usr/share/vosk/models/$model
is a directory, if so use it.
NOTE: Vosk-cli requires an exact match between directory and model name. If you pass -m eng
you must have an eng
model in your model search path!