Sound-Classification-Mel-Spectrogram

Definition

This project classifies sound signals from different environmental classes in the ESC-10 dataset. the above photo summarizes the model steps:

1. The model read all the signals of different classes and assign a label number to each class.
2. The Mel Spectrogram are extracted from the time domain.
3. Full Convolutional Neural Network(CNN) is defined and used to classify 10 different classes of ESC-10 dataset.

Dependencies

To run this project you will need to:

1. download the ESC-10 dataset from this link: ESC-10

2. change the directory name that contains the dataset to the name in the notebook file in these three lines:
   

#here my directory name is "dataset"

data, samplerate = librosa.load("dataset/dog/1-30344-A.wav", sr=44000) 

for filepath in glob.iglob('dataset/*'):

for j in glob.iglob('dataset/'+i+'/*'):

install this libraries:

• numpy
• keras
• matplotlib
• librosa
• pylab
• glob
• tensorflow