Bachelor's Thesis - by Patrick Gerber and Sebastian Glinski-Haefeli
src/core/spectrogram_generation/main_data_extractor.py
: Extract mel-spectrograms from speaker audio files and save them todata/training/TIMIT_extracted
in .pickle file format
src/network_runner.py
: Wrapper functions to start training a model with appropriate parameters. In this case it's the networksrc/nets/bilstm_2layer_dropout_plus_2dense.py
regarding the results from the bachelor's thesis ba17_stdm_1.
src/keras_cluster_output_generator.py
: Execute clustering procedure on appropriate extracted speaker data with the previously trained network (-model)src/clustering/cluster_tester.py
: Calculate missclassification rate and optionally show a scatter plot for processed clusterings
- data/experiments/logs/
- data/experiments/nets/
- data/experiments/plots/
- data/speaker_lists/
- data/training/TIMIT/
- data/training/TIMIT_extracted/