Classifier Comparison for Blind Source Separation

Authors:
Publisher website
Google Scholar
Abstract
Source separation is important task in intelligibility for human as well as speech recognition system. The command from speech is required to be recognized for voice operated Internet of Things devices. The speech separation is complex task when number of noise sources increase with respect to room acoustics. The mixing of sounds and their separation is experimentation approach considered. The feature extraction is performed using 1D CNN model. The very less complex CNN model provides various in depth features which can be easily separated in the classifier. The model is proposed using recurrent neural network (RNN). The performance is evaluated and compared with other machine learning classifiers for classification of commands. The accuracy of more than 98% is seen with the use of proposed RNN model with combination of gated recurrent unit (GRU) and long short term memory (LSTM).

This publication has 18 references indexed in Scilit: