dscottboggs/vosk-script

The script I use to transcribe stuff with vosk

STT script

This is a a speech-to-text script. It's very simple and doesn't do much.

Requirements

Minimum

• around 500MiB RAM
• 40-50MiB disk space for the "small" model (assuming your locale has one), not including your source media.

Recommended

• 8GiB RAM (the full english model hits OOM on my 4GiB laptop)
• 2GiB disk space. Some locales are smaller but they are all around 1-2GiB for the full-size model.

Installation

• Install ffmpeg
• Install python.
• Install vosk:

$ pip install vosk

• Download this script

  It can go anywhere, just run it like python /full/path/to/vosk-script.py. However, you can also place it in a folder on your $PATH, such as ~/.local/bin, and make it executable, then you will be able to run just vosk-script.py from anywhere. Like so:

$ curl https://raw.githubusercontent.com/dscottboggs/vosk-script/master/vosk-script.py > ~/.local/bin/vosk-script.py
$ chmod 755 ~/.local/bin/vosk-script.py

• Download and unzip a model from this page. For example:

$ mkdir ~/.local/share/vosk-models
$ cd ~/.local/share/vosk-models
$ wget https://alphacephei.com/vosk/models/vosk-model-en-us-0.22.zip
$ unzip vosk-model-en-us-0.22.zip
$ ln -s vosk-model-en-us-0.22 english

• Run the script. Run ./vosk-script.py --help for more information.