Overview of Machine Learners in Classifying of Speech Signals

Overview of Machine Learners in Classifying of Speech Signals

Hemanta Kumar Palo (Siksha O Anusandhan(Deemed), India) and Lokanath Sarangi (College of Engineering, Biju Patnaik University of Technology, India)
DOI: 10.4018/978-1-5225-9643-1.ch022

Abstract

Machine learning (ML) remains a buzzword during the last few decades due to the requirement of a huge amount of data for adequate processing, the continuously surfacing of better innovative and efficient algorithms, and the advent of powerful computers with enormous computation power. The ML algorithms are mostly based on data mining, clustering, classification, and regression approaches for efficient utilization. Many vivid application domains in the field of speech and image signal processing, market forecast, biomedical signal processing, robotics, trend analysis of data, banking and finance sectors, etc. benefits from such techniques. Among these modules, the classification of speech and speaker identification has been a predominant area of research as it has been alone medium of communication via phone. This has made the author to provide an overview of a few state-of-art ML algorithms, their advantages and limitations, including the advancement to enhance the application domain in this field.
Chapter Preview
Top

Introduction

Use of data-driven approach in ML makes it more reliable than man-made rules. It is more dependable in a situation where a human being remains inefficient to cope with. It follows an automated method to find the desired hypotheses that explain the data without any human experts. The ML algorithms can be applied to almost any learning tasks with flexibility and economically. A few applications of the ML algorithm are shown in Figure 1.

Figure 1.

Applications of machine learning algorithms

978-1-5225-9643-1.ch022.f01

Advancement in classification algorithms involving speech signal can create a suitable platform for resource allocation, gender and age detection, language identification, digit and word recognition, emotion identification, security and criminal investigation, robotics and computer games, on-line tutoring, counseling and psychological assistance to children and affected person and many similar fields (Sinha & Shahnawazuddin 2018; Palo & Mohanty 2018A; Palo, Chandra & Mohanty, 2017A; Geiger, Leykauf, Rehrl, Wallhoff, & Rigoll, 2014;Franke&Srihari 2008; Litman& Forbes 2003). The recognition system allows a machine to emphasize on phrases or words in a spoken conversation. Similarly, speech to text conversion helps the user to verbally communicate with the machine rather than typing the text chosen. Thus, the ML algorithm must be trained with numerous amounts of words, vocabularies, terminologies (medical, industrial, technical, legal, etc.), phrases, and languages to make it user-adaptive. A few of the other real-world applications are listed below

  • Dialogue management: The human-machine-interface must include dialogues to enable either the machine or the user to initiate or choose different responses to genuine queries. However, sufficient progress in this field is yet to be made to provide such mixed-initiative recognition systems.

  • Telecommunications: Credit card recognition, third party billing, operator-assisted calling, rejection or acceptance of billing charges identifying the selected speaker, voice dialing (call the workplace, school or home, etc.) call-center application, voice calling, customer care, command and control of resource allocation, etc.

  • Desktop/office management: Internet voice browsing, voice dialer, desktop voice navigation, dictation, etc.

  • Legal/ medical: The creation of medical and legal reports using speech to text conversation.

  • Aids to handicapped/ Games/ robotics: Manipulation of wheelchair carrying patients, climate control, control of selective parameters of games using speech signal, human-robot interaction, etc.

  • In-car application: Audio prompting the driver by identifying the known passenger, vehicular maneuvering, initiate phone calls, tuning the radio stations or activating music systems, etc.

  • Military applications: Setting up the radio-frequencies, steer-point, autopilot commanding, coordinate and control of weapon release features, control of flight display system, etc.

  • Education system: Language learning (choice of proper pronunciation, enhancing speech fluency and speaking style or skills), voice activation system and information dissemination to a blind person, relieving a paralyzed person from writing, typing, browsing of the internet or use of computers using their recognized voice samples, etc.

Complete Chapter List

Search this Book:
Reset