- extractFeatures.py: extracts features from audio signal
- generateDatasets.py: loops through specified dir's to generate training and test datasets then saves datasets to files for later use
- neuralNetwork.py: sets up a NN, loads the saved datasets to use for training, runs a training session then runs a testing session. Outputs accuracy score, a plot of training epoch vs. cost, a confusion matrix, and final score.
-
use ffmpeg to cut 4sec. segments from audio files.
-
example in MS-DOS:
ffmpeg -i C:\users\joe\file -ss 00:00:45.0 -t 4 -acodec copy chimp1.flac
-
-
set paths in generateDatasets.py lines 8-11 to directories containing segmented audio files.
-
set paths in generateDatasets.py lines 66-69 to desired save destination for generated datasets.
-
set paths in neuralNetwork.py lines 13-16 to files containing saved datasets which were generated in Step 3.
-
Line 40 of neuralNetwork.py contains code to produce visual representation of dataflow graph using Tensorboard. Set the event file desitination relative to your machine. Then in CLI call two commands:
activate Tensorflow
Tensorboard --logdir=directory_containing_event_file