gender recognize using pre-processed voice data by deep neural networks
- tensorflow (for tensorboard logging)
- pytorch (>=1.0, 1.0.1 used in my experiment)
- pandas and other common python packages
from kaggle and excel files are also attached in this repo
This repo contains two structures of neural networks to recognize gender of pre-processed voice data (2-classification problem)
run python with_fully_connect.py
to train a model using 3 fully connected layers, and finally a model parameters file would be saved. It is 2.07MB in size
and the accuracy in training set would approch nearly 100% and 96% in testing set.
run python with_conv1d.py to train a model using conv1d with much more layers depth than the fully connected, and the residual learning strategy is used to handle the deeper depth training.
finally a model parameters file would be saved and it is 57.2MB in size (much bigger than fully connected layers model)
the accuracy in training set would approch nearly 100% and 97% in testing set (quite slight improvement~ but it works).
AFTER TRAINING, you can run python test_model.py -f and python test_model.py -c to test the trained fully connected model and conv1d model in test set respectively,
and it would produce a txt file contains the female probabilities of test data per row.