Audio and Speech Processing with MATLAB

Paul Hill

• Publisher: CRC Press
• Year: 2018
• eBook ISBN: 9780429813955
• Edition: 1
• Topic: Computer Science
• Subtopic: Computer Networking
• Index: Computer Science

Speech and audio processing has undergone a revolution in preceding decades that has accelerated in the last few years generating game-changing technologies such as truly successful speech recognition systems; a goal that had remained out of reach until very recently. This book gives the reader a comprehensive overview of such contemporary speech and audio processing techniques with an emphasis on practical implementations and illustrations using MATLAB code. Core concepts are firstly covered giving an introduction to the physics of audio and vibration together with their representations using complex numbers, Z transforms and frequency analysis transforms such as the FFT.

Later chapters give a description of the human auditory system and the fundamentals of psychoacoustics. Insights, results, and analyses given in these chapters are subsequently used as the basis of understanding of the middle section of the book covering: wideband audio compression (MP3 audio etc.), speech recognition and speech coding.

The final chapter covers musical synthesis and applications describing methods such as (and giving MATLAB examples of) AM, FM and ring modulation techniques. This chapter gives a final example of the use of time-frequency modification to implement a so-called phase vocoder for time stretching (in MATLAB).

Features

• 
  
  

• A comprehensive overview of contemporary speech and audio processing techniques from perceptual and physical acoustic models to a thorough background in relevant digital signal processing techniques together with an exploration of speech and audio applications.

• 
  
  

• A carefully paced progression of complexity of the described methods; building, in many cases, from first principles.

• 
  
  

• Speech and wideband audio coding together with a description of associated standardised codecs (e.g. MP3, AAC and GSM).

• 
  
  

• Speech recognition: Feature extraction (e.g. MFCC features), Hidden Markov Models (HMMs) and deep learning techniques such as Long Short-Time Memory (LSTM) methods.

• 
  
  

• Book and computer-based problems at the end of each chapter.

• 
  
  

• Contains numerous real-world examples backed up by many MATLAB functions and code.

1

MATLAB^® and Audio

CONTENTS

1.1 Reading Sounds

1.2 Audio Display and Playback

1.3 Audio-Related MATLAB

1.4 Example Audio Manipulation

1.5 Summary

1.6 Exercises

To me, it’s always a joy to create music no matter what it takes to actually get there. The real evils are always whatever stops you from doing that – like if your CPU is spiking and you have to sit there and bounce all your MIDI to audio. Now that’s annoying!

Skrillex

MATLAB (MATrix LABoratory) is used throughout this book for audio processing and manipulation together with associated visualisations. This chapter therefore gives an introduction to the basic capabilities of MATLAB for audio processing. Appendix B also gives a list of core MATLAB functions and commands that are applicable to audio processing for those starting to use the language. This chapter (and Appendix B) can be skipped or skimmed if the reader is familiar with the basic operations of the MATLAB programming language and its visualisation capabilities.

More general information is given on the Mathworks (creators of MATLAB) website www.mathworks.com and within the innumerable help files, demos and manuals packaged with MATLAB.

TABLE 1.1: Audio formats available to be read by MATLAB command audioread

Audio File Format	Description	File extension
WAVE	Raw audio	.wav
OGG	OGG vorbis	.ogg
FLAC	Lossless audio compression	.flac
AU	Raw audio	.au
AIFF	Raw audio	.aiff,.aif
AIFC	Raw audio	.aifc
MP3	MPEG1 Layer 3, lossy compressed audio	.mp3
MPEG4 AAC	MPEG4, lossy compressed audio	.m4a, .mp4

1.1 Reading Sounds

Audio is read into MATLAB using the function audioread whose basic functionality is as follows.¹

Where filename in this case is a MATLAB variable containing a string (array of chars) defining the entire name of the audio file to be read including any file extension (e.g., mp3, wav, etc.). A typical example of a call to audioread would be

where y is the array or matrix of sampled audio data and Fs is the sampling frequency of the input audio. audioread is able to read the formats shown in Table 1.1. In this example, filename is ‘exampleAudio.wav’, the file to be read in (filename is required to be of a MATLAB string type and is therefore delimited by single quotes ’). filename can be a MATLAB string that can also include a path (defined in the format of your operating system) to any location on your hard drive. For example, filename could be ‘c:\mydirectory\mysubdirectory\exampleAudio.wav’ (on windows) or ‘~/mydirectory/mysubdirectory/exampleAudio.wav’ (on OSX/Unix/Linux). A statement in MATLAB will automatically display its results. It is therefore common to want to suppress this output and this is achieved by using the semicolon at the end of each line where no output is required.

It is often useful to determine detailed information about an audio file before (or indee...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright Page
• Dedication
• Table of Contents
• Preface
• List of Acroynms
• Introduction
• 1 MATLAB® and Audio
• 2 Core Concepts
• 3 Frequency Analysis for Audio
• 4 Acoustics
• 5 The Auditory System
• 6 Fundamentals of Psychoacoustics
• 7 Audio Compression
• 8 Automatic Speech Recognition: ASR
• 9 Audio Features for Automatic Speech Recognition and Audio Analysis
• 10 HMMs, GMMs and Deep Neural Networks for ASR
• 11 Speech Coding
• 12 Musical Applications
• A The Initial History of Complex Numbers
• B MATLAB Fundamentals (Applicable to Audio Processing)
• Index