Jupyter notebooks for the project "Gender recognition by voice" for the EPFL Statistical learning course. The objective is to predict whether a voice recording is 'male' or 'female' based on spectral features extracted from the each recording. The dataset, available on Kaggle, is a '.csv' file which consists of features extracted 3248 voice recordings.

1. Install Python 3.6 and Anaconda.
2. Create an anaconda environment.

   conda create -n gender_recognition python=3.6
   source activate

3. Clone or download the repository.

   git clone https://github.com/AdriBesson/Gender_recoginition_speech
   cd Gender_recoginition_speech

4. Install the python dependencies

   pip install --upgrade pip
   pip install -r requirements.txt

1. Run the different notebooks: the notebooks follow the process described in ''Report_Besson_Lefebvre_Kaklamanos.pdf'':

• Gender recognition by voice - data analysis.ipynb: Exploratory data analysis described in Chapter 3
• Gender recognition by voice - A first intuitive approach.ipynb: Classification using only the fundamental frequency described in Chapter 4
• Gender recognition by voice - Classification with 80-20 split.ipynb: Classification using all the features and a 80/20 split of the dataset. Training set is used for parameter selection test set used for model comparison. Due to a high variance of the classification error with respect to the split of the dataset, this approach is not able to identify the best model. Described in Chapter 4
• Gender recognition by voice - Classification with 50-50 split.ipynb: Classification using all the features and a 50/50 split of the dataset. Training set is used for parameter selection test set used for model comparison. To reduce the variance, we use 5-fold cross validated scores on the test set. Described in Chapter 5