Wakeup Word Detection & Speaker Recognition

A tensorflow2.x implementatino of Wake-up Word Detection and Sepeaker Recognition

• Detect Wake-up Word
• Recognize Speaker
• tflite convert (Just type, quantization, pruning)

Environment Setting

1. Download requeirments.txt file
2. Edit configuration.py

dataset

1. Collect your data in custum_dataset folder for each users

Data setting

1. change_name.py
2. data_augmentation.py (reverse, stretch, pitch, shift, noise)

Audio to Image(Mel-log Spectrogram)

Make Mel-log Spectrogram & data preprocess (Regularization, clip etc)

1. *_data_preprocess.py (generate .npz file for train, valid, and test)

Train and evaluate

Best model will be saved in ./best_model

1. *_train.py
2. test_all.py

Convert tensorflow to tflite

1. orig_tflite.py
2. quantization_tflite.py
3. pruning_tflite.py
4. tflite_evaluate.py ( evaluate all tflite model )

Real time detectino demo

Check demo sample in ./demo

1. python realtime_wuw.opy