HMG is a GUI tool designed to train, evaluate and export keyword spotting models.
DISCLAIMER: This is an rough version of a tool that haven't beed thoroughly tested, bug may (and probably will) occur. Please use the issue tracker to report them.
HMG allows you to:
- Manage datasets
- Tune audio features
- Tune model architecture and training parameters
- Train your model
- Evaluate your model
- Test your model
- Export your model
Not available for now
You can use the software from the repository.
Prerequisites
System dependencies:
- You need
python3.x
andpython3-pip
installed. - you need
portaudio19-dev
package installed
There is a number of dependencies needed to run the software.
They can be found in the requirements.txt
file at the repository root.
We recommend that you use a virtual environement.
source /my/virtual/env/bin/activate #Optionnal but advised
cd $(REPO_ROOT)
pip install -r requirements.txt
You can launch the program by running either:
python $(REPO_ROOT)/model_generator/main.py
# or
./$(REPO_ROOT)/model_generator/main.py
- Create a project and specify your keyword(s).
- Create a dataset and import samples
- Create a features profile
- Create a model architecture
- Create a trained model and train it
- Evaluate your model
- Export your model
- Qt - Cross-platform software development for embedded GUI
- PyQt - Python binding for Qt
- Tensorflow - An open source machine learning library for research and production
This project is licensed under the GNU AFFERO License - see the LICENSE.md file for details.