Mmeslay

• An Automatic Speech Recognition System for the Kabyle language with a flutter Frontend.

• This was created by training the Squeezeformer-XS model using the Common Voice (the subset in kabyle language).

• The model was trained, validated and tested on a custom split of the dataset.

• A language model was also trained on a text corpus composed of sentences collected from various sources, such as Tatoeba and https://github.com/MohammedBelkacem/Kabyletexts, using KenLM.

• The system was tested using various configuration of the CTC decoder :

  

Prerequisites

• Run pip install Cython then pip install -r requirements.txt to install the dependencies. Using Conda or something similar is recommeneded.
• Run flutter pub get in the root of the frontend folder to install the required dependencies.
• The app was tested on Python 3.9, Flutter 3.10 and Dart 3.0.

How to use

• To start the backend, cd into the backend folder and run python src/flask_server.py
• Run flutter pub get in the root of the frontend folder to install the required dependencies.
• Build the front end from source or install the appended release .apk.
• The front end was only tested on Android, although, it should be able to run on other platforms supported by flutter, but it may need some tweaks
• The default ip adress for both the frontend and the backend is 192.168.12.1, you can change it in the source code to suit your needs.

Cite us

• If you use our project in your work, please cite us :

@mastersthesis{Mmeslay,

  author  = {Aomer Gaya Ouldali},

  school  = {Université A. Mira de Béjaïa},

  title   = {Système de reconnaissance de la parole appliqué à la langue Tamazight},

  year    = {2023}

}