This is a python package containing the Command Line Interface for the radioship_transcriber.
The transcriber creates transcripts for .mp3 files in .txt format using neural networks created for this purpose. It takes an input folder path, an output folder path, and an optional model path that can be a local path or a url on huggingface.co. If no model path is given, it will use our hungarian model as a default.
The transcriber is dependent on the huggingsoud package, that uses a specific range of versions of torch. Since older versions of torch are not compatible with the latest versions of python it is important to use python 3.10.10. We recommend using pyenv to install it. Here is a great tutorial for that.
You will also need ffmpeg to process .mp3 files. If you don't have it on your machine:
sudo apt-get install ffmpeg
Then create a new virtual environment with pipenv:
pipenv install --python 3.10.10
Activate the new env with:
pipenv shell
Make sure it really uses python 3.10.10! Then install the radioship_transcriber from this repo:
pipenv install git+https://github.com/Koffair/radioship_transcriber.git#egg=radioship_transcriber
Now, if your virtual environment is active, you can call the transcriber the following ways (no need to type python):
-
to use default model:
radioship_transcriber -i path/to/input/ -o path/to/output/ -
to use your own or any other model:
radioship_transcriber -i path/to/input/ -o path/to/output/ -m path/to/other/model/
The input folder should contain .mp3 files. The output folder will have .txt files, and the logs in a separate folder.
If the specified model is not present, the transcriber will download and cash it for later use.
If you want to remove a cashed model, use:
huggingface-cli delete-cache
Create transript for mp3 files. [-h] -i -o [-m]
options:
-h, --help show this help message and exit
-i , --in_path Path to input file or directory
-o , --out_path Path to output directory
-m , --model_path Address to transcripter model