gender recognize using pre-processed voice data by deep neural networks
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 finnaly 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.
finnaly 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.
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。