Audio Classification
DESCRIPTION
• In this model the audio was classfied using Convolution Neural Network algorithm.
• PUBG game audio was used as a datasets to classify the sound based on object detection using TensorFlow coco model (ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8)..
METHODOLOGY
PRE-PROCESSING
• PUBG game audio was recorded in mp3 format and then converted into wav format for the purpose of converting the audio into mel-spectrogram frequency image.
• Librosa and SciPy both is a open source libraries supports wav file for both read and write operation but it’s not feasible in mp3 format.
• By using Librosa, Matplotlib and SciPy library, we were able to extract the Mel-spectrogram and save it as images.
MEL-SPECTROGRAM OUTPUT
Gun Change Audio
Gun Shot Audio
M416 Gun Load Audio
M762 Gun Load Audio
Pistol Gun Load Audio
DATA AUGUMENTATION
• Data augmentation is labeling the image for classification.
• LabelImg tool was used for labeling the image which will generate a XML file for bound box annotation.
• The XML code are generate from labelImg tool for labeling the images and it’s divided into two part: One is for training and other is for testing.
MODEL TRAINING AND TESTING
• Model was trained using Tensorflow coco ssd_mobilenet_v2_fpnlite_640x640_coco17_tpu-8 model.
MODEL OUTPUT
Gun Change Output
Gun Shot Output
M416 Gun Load Output
M762 Gun Load Output
