Automatic Speech Recognition: Simple calculator

How to start

git clone https://github.com/pmikolajczyk41/Automatic-Speech-Recognition-Simple-calculator.git
cd Automatic-Speech-Recognition-Simple-calculator/

python3 -m venv ./venv
pip install --upgrade pip
pip install -r requirements.txt

export PYTHONPATH=`pwd`

Data format

There are 15 base models: one for silence, one for each digit and one for each operator:

• "+" (plus)
• "-" (minus)
• "*" (times)
• "/" (by)

To train them you need to provide .wav files to resources/train/ in a format <name>-<id>.wav where <name> corresponds to a symbol and <id> is a unique number for distinguishing between different recordings.

How to train

The main section within calculator/atom_training.py script is responsible for training atom models. After providing training data to the resources/train/ directory you are ready.

python data/convert.py
cd calculator/
python atom_training.py

Notice that by default Viterbi training preceded by a uniform segmentation is applied. To retrain you models just comment/uncomment appropriate lines in calculator/atom_training.

How to test

cd calculator/
python testing.py

How to run app

cd calculator/
python app.py

You may get some errors connected with audio setup within your system. Here is a sample common problem raised with some solutions.