This has only been tested and is only expected to work using Python 3.6 [64 bit] due to tensorflow requirement.
A package to perform Speech/Non-speech Identification (SNI) using Neural Networks on .wav files.
The function that performs SNI is in VAD.py and is called Neural_Network_VAD. Provided the speech signal (as a 1-D array) and the sampling frequency, it returns as a tuple the SNI results from prediction using a Convolution-LSTM-Dense Neural Network (0 index of tuple) and a LSTM-Dense Neural Network (1st index of tuple).
VAD_script.py is a wrapper that will save the results of SNI as a plot in a .png file and as a csv file.
use: python VAD_script.py sample.wav
for help: python VAD_script.py -h
A sample wav file has been include in which the background noise has been obtained from [1] and the speech from [2].
[1] Koenig, M. (2018). Street Sounds | Effects | Sound Bites | Sound Clips from SoundBible.com. [online] Soundbible.com. Available at: http://soundbible.com/2175-Street.html [Accessed 14 Jun. 2018].
[2] Fromtexttospeech.com. (2018). From Text To Speech - Free online TTS service. [online] Available at: http://www.fromtexttospeech.com/ [Accessed 14 Jun. 2018].