a simple neural network with LSTM node, that recognize voices genders.
I use MFCC and etc features for this task, u can see feature_extraction.py
read libs.txt file, and install tensorflow and keras too.
for running with my train and test extracted features :
python train.py
for extraction your own vioces put your train *.wav
files into this folders:
├── data
│ ├── test
│ │ ├── female
│ │ └── male
│ └── train
│ ├── female
│ └── male
then for extract features of your voices (it put them in saved_features
folder) run:
python save_features.py
then for train, run:
python train.py
train accuracy => ~100%
test accuracy => ~88%
train_male samples count => 121
train_female samples count => 121
test_male samples count => 36
test_female samples count => 36
LSTM nodes count => 20
epochs => 500
any better idea? please open an issue, thanks.