Skip to content

Latest commit

 

History

History
46 lines (39 loc) · 1.39 KB

Readme.md

File metadata and controls

46 lines (39 loc) · 1.39 KB

Speakato Trainer

Speakato Trainer is a project used to create models conforming to the requirements of Speakato.

Write down the desired commands, prepare dataset and Speakato will do the rest.

WARNING: Only Windows is supported!

Prerequisites

Before you start don't forget to install necessary packages:

pip install -r requirements.txt
python -m spacy download pl_core_news_sm
python -m spacy download en_core_web_sm

...or use setup.cmd to configure virutal environment with all required packages and modules!

How to build training dataset?

Put the list of commands along with it's labels in the dataset.json. Data should be saved in the following form:

[
    {
        "text": "some text",
        "command": "example_command"
    }
]

Save command list in the commands.txt file:

greeting
open
close
alarm

Sample datasets have been included in examples\

How to use?

Prepare dataset, run python src\main.py and pass necessary arguments. That's all!

Example usage with flags: python src/main.py --language pl --dataset .\examples\polish_commands_dataset\ --model .\models\test --mode 1

Available modes:

  • 1 - Create new model
  • 2 - Add new data to previously created model

How to test trained model?

You can use predefined methods from src\Sandbox.ipynb to test your model.