treinaPB
This project is under GNU General Public License v3.0.
Dependencies
The program was tested in a GNU/Linux operating system (Debian 11) using the following packages:
- Python 3.9
- ffmpeg 4.3.3
- Hidden Markov Model Toolkit (HTK) 3.4.1
- Praat 6.2.0.6
If you don't have Python 3 or ffmpeg, install it. In Debian-based GNU/Linux:
apt install python3 ffmpeg
The program depends on the following Python packages:
- pydub 0.25
- pyenchant 3.2.0
- sly 0.4
- nltk 3.6.7
- termcolor 1.1
- pandas 1.4
- numpy 1.12.0
- pyyaml 4.2b1
- scipy 0.18.1
- TextGrid 1.4
To automatically install these packages, go to the project's main folder and run in the console:
pip3 install -r requirements.txt
The program also depends on the punkt NLTK data. To install it directly in the command line, run:
python3 -m nltk.downloader punkt
Finally, a pt_BR (Brazilian Portuguese) dictionary also have to be installed in your computer. You can install it from one of these two providers: hunspell or aspell.
If you're running a Debian-based distribution of GNU/Linux, it should be available on the apt repositories:
sudo apt install hunspell-pt-br
How to run?
- Clone this repository
- Install all the dependencies
- Open a terminal emulator and run:
python3 run.py -d <directory> -r <reference tier> -i <tiers to ignore>
Help
usage: TreinaPB [-h] [-d DIRECTORY] [-r REFERENCE] [-i [IGNORE ...]]
Training HTK models for forced alignment in Brazilian Portuguese.
optional arguments:
-h, --help show this help message and exit
required arguments:
-d DIRECTORY, --directory DIRECTORY
path to the directory with recording (.wav) and
trascription files (.eaf).
-r REFERENCE, --reference REFERENCE
name of the reference tier in the transcription files.
-i [IGNORE ...], --ignore [IGNORE ...]
names of the tiers to ignore when checking for
overlaps.