A Preprocessing methodology for improvising Speech Systems

Abstract

Machine learning is the ability of a computer to learn to make decisions without being programmed explicitly. It focuses on making such computer programs that have the ability to change when in contact with new data. Data prediction and learning from data is the main task of such programs and machine learning helps to explore the study and construction of algorithms which help to do the same. We are witnessing rapid advancements in this field and it is a possibility in the near future that voice interaction systems will replace our normal way of interaction through standard keyboards in the near future. Today, some notable examples in voice interaction systems are Microsoft Kinect and Apple SIRI which perform really well, but, like any other technology, there is a wide scope of improvement in each of the speech systems which are in existence. There are limitations and research is carried out to improvise the effectiveness of these types of systems. Gender recognition which is one of the preprocessing methodologies can be utilized to improvise the performance of speech systems. In this paper, automatic gender identification system using acoustic properties of speech is discussed. Further, usage of acoustic analysis to process different samples is carried out and then applied to different types of machine learning algorithms which help to learn gender-specific traits. The fore-mentioned topic is being discussed and experimented.