tensorflowasr-scripts

This repository stores useful scripts for translating between transcript types with sclite and the switchboard corpus for TensorFlowASR.

Setup

This repository is for Python 3, tested on 3.7.3

python3 -m pip install -r requirements.txt or pip3 install -r requirements.txt

The switchboard corpus contains sph (sphere) files instead of WAV files, so sphfile is used to convert them.

Scripts

create_switchboard_trans.py

This script is used for creating a .tsv file from the switchboard's .trans file for use in TensorFlowASR.

There is a chance that your transcript will be different, and you would then need to change the split value and index accessed. Feel free to create an issue if you run into that issue.

Arguments:

--utt Used to specify the directory that contains the .sph files.
--trans Used to specify the .trans transcript file to be converted.
Unnamed argument: The output .tsv file

Example:

python3 ~/capstone/repos/tensorflowasr-scripts/create_switchboard_trans.py --utt ../audio/utt/ --trans train.trans train.tsv`

parse_tsv.py

This script is used to convert the .tsv to a .trans file for decoding with sclite.