Mossaida is an APP powered by NLP (Natural Language Processing) to help people with reduced mobility in Human-Machine interactions. For the moment these application is just available for french langage.
In a nutshell, the app is able to get the user voice and execute its commands.
➡️ List of possible commands :
- créer des dossiers/fichiers
- supprimer des dossiers / fichiers
- ouvrir des dossiers /fichiers
- fermer des dossiers /fichiers
- éditer un fichier
- lire un fichier
- copier un fichier/dossier dans un autre dossier
- augmenter/ diminuer le volume
- augmenter / diminuer la luminosité de l’écran
To achieve our goal, we build a NER model with Spacy on our own dataset and Entities (ACTION, CIBLE, VALEUR). The details about model training are in the NER folder.
To tag the dataset with our own entities we use NER ANNOTATOR and the results are stored in the file annotations.json
- Install portaudio
sudo apt install portaudio19-dev
- Clone this repository
git clone https://github.com/LiganiumInc/Mossaida.git
cd Mossaida
- Create a virtual environment and install required libraries
python -m venv env
source env/bin/activate
pip install -r requirements.txt
- Download the Vosk model for Speech-to-Text here and unzip it in this directory
- Run the app
streamlit run app.py
A web interface will be open in your browser on http://localhost:8501 and your microphone will be activate. Then, speak some commands like those in file data_train.txt (see NER folder). You can look logs in real time in your terminal.