This project trains a GMM (Gaussian Mixture Model) to identify speaker. Model is trained and evaluated on single-speaker recordings in (ideally) controlled environments. Multiple preprocessing and feature extraction techniques are used.
- Python 3.x
- pip
Install pip packages using
$ pip install -r requirements.txt
Then put your local dataset into data file in project root. Your dataset should be split into 3 subfolders, train, test and eval. Train and test folders should have a subfolder for each speaker with his/hers id as folder name containing {whatever}.wav files with recordings. Eval folder is flat and should contain wav files directly.
Example:
.
└── data
├── train
│ ├── speaker1
│ │ ├── a1.wav
│ │ ├── a2.wav
│ ├── speaker2
│ │ ├── b1.wav
│ │ ├── b2.wav
│ └── ...
├── test
│ ├── speaker1
│ │ ├── a1.wav
│ │ ├── a2.wav
│ ├── speaker2
│ │ ├── b1.wav
│ │ ├── b2.wav
│ └── ...
└── eval
├── unknown1.wav
└── ...
Train GMM model using command
$ python src/gmm_train.py
After training the gmm params are saved to model/gmm/ folder and processed training dataset to data/train.npy.
Evaluate using command
$ python src/gmm_eval.py
🎉 🎉