1. Week 8: Audio Processing 1

2.  Light and sound are both transmitted in waves 2

3.  The human ear can hear between about 12 Hz and 20,000 Hz  The higher the frequency of the wave, the higher the frequency of the note  Note that the A an octave above concert A (A440) has twice the frequency  Each half-step is an increase in the frequency by a factor of about 1.06 NoteFrequency A440 B493.88 C523.25 D587.33 E659.26 F698.46 G783.99 A880 3

4.  We can take a sound:  And reproduce that sound at double the frequency (speeding it up):  Notice that we have to add twice as much information to have the sound fill the same amount of time 4

5.  The amplitude of a wave is the distance to its peak (measured by its y-value)  In sound, amplitude is a measure of volume  The larger the amplitude, the louder the sound 5

6.  We can take a sound:  And make the sound with half the amplitude:  The frequency is exactly the same, but the sound is half as loud 6

7.  Something that looks like a smooth sine wave is called a pure tone  No real instruments play anything like that  Even the purest real sound has overtones and harmonics  Real sound is the result of many messy waves added together: 7

8.  On a computer, we cannot record a wave form directly  As usual, we have to figure out a way to store a wave as a series of numbers  We are going to use these numbers to approximate the heights of the wave at various points 8

9.  Hertz (Hz) is a unit that means a number of times per second  We are going to break down the wave into lots of slices  We are going to have 44,100 slices in a second  Thus, we are slicing at 44,100 Hz 9

10.  We slice up a wave and record the height of the wave  Each height value is called a sample  By getting 44,100 samples per second, we get a pretty accurate picture of the wave 10

11.  There are many different formats for sampling audio  In our system, each sample will be recorded as a double  The minimum value of a sample will be -1.0 and the maximum value of a sample is 1.0  A series of samples with value 0.0 represents silence  Our samples will be stored in an array 11

12.  Audio data on computers is sometimes stored in a WAV file  A WAV file is much simpler than an MP3 because it has no compression  Even so, it contains two channels (for stereo) and can have many different sample rates and formats for recording sound  The StdAudio class lets you read and write a WAV file easily and always deal with a single array of sound, sampled at 44,100 Hz 12

13.  Everything you’d want to do with sound:  To do interesting things, you have to manipulate the array of samples  Make sure you have StdAudio.java in your directory before trying to use it MethodUse static double[] read(String file) Read a WAV file into an array of double s static void save(String file, double[] input) Save an array of double s (samples) into a WAV file static void play(String file) Play a WAV file static void play(double[] input) Play an array of double s (samples) 13

14.  Let’s load a file into an array:  If the song has these samples:  Perhaps sample will contain: String file = “song.wav”; double[] sample = StdAudio.read(file); -.9-.7-.6-.4-.2-.1.1.2.3.4.5.6.5.4.3.20-.2-.4 14

15.  With the audio samples loaded into the array named sample, we can play them as follows: StdAudio.play(sample); 15

16.  Or, we could generate sound from scratch with StdAudio  This example from the book creates 1 second of the pitch A440:  StdAudio.SAMPLE_RATE is 44100 double[] sound = new double[StdAudio.SAMPLE_RATE + 1]; for( int i = 0; i < sound.length; i++ ) sound[i] = Math.sin(2 * Math.PI * i * 440 / StdAudio.SAMPLE_RATE); StdAudio.play(sound); 16

17.  The book provides a short program called Play That Tune (pp. 150 and 205) that will generate a sequence of notes according to an input file  It’s a sort of software synthesizer  If you know the notes for a song you’d like to play and their durations, you can create a file to play it with PlayThatTune.java  java PlayThatTune < song.txt 17

18.  As you can see, the file is just a list of numbers giving pitch and duration  Although ugly, this form of bookkeeping takes up virtually no space compared to WAV files  Drawbacks:  You can only play one note at a time  All notes have a pure tone with a little bit of harmonics  It takes forever to type out an input file  No dynamics (volume control) 18

19.  Remember, sound is a wave  We are going to start playing with the waveform now to get different effects  Phase is a property of a wave that can be determined by its starting position  You can think of phase as the alignment of the wave 19

20.  What happens if we invert a sound wave?  For example, we turn this wave:  Into this wave: 20

21.  How would we invert the phase of a wave in code?  And what does that crazy, upside-down sound sound like?  Exactly the same  The human ear is not sensitive to phase double[] sound = StdAudio.read( file ); for( int i = 0; i < sound.length; i++ ) sound[i] = -sound[i]; StdAudio.play(sound); 21

22.  Phase is still worth knowing about  If you hear two copies of the same sound, but one is 180° out of phase, the sounds will cancel each other out  This can happen if you hook up one of your stereo speakers backwards 22

23.  To reverse a sound, we simply send the wave form backwards  For example, we take this sound:  And turn it into this sound: 23

24.  How would we reverse a sound in code?  What does it sound like?  Sort of like a bizarre foreign language double[] sound = StdAudio.read( file ); int start = 0; int end = sound.length - 1; while( start < end ) { double temp = sound[start]; sound[start] = sound[end]; sound[end] = temp; start++; end--; } StdAudio.play(sound); 24

25.  Recall that the amplitude of a wave is the distance to its peak (measured by its y-value)  In sound, amplitude is a measure of volume  The larger the amplitude, the louder the sound 25

26.  We can take a sound:  And make the same sound with half the amplitude:  The frequency is exactly the same, but the sound is half as loud 26

27.  How would we half the volume of a sound in code?  This is half the amplitude. Is this exactly half the volume?  No… this stuff gets complicated  Different frequencies with the same amplitude are actually perceived as different loudnesses  But, it’s close enough double[] sound = StdAudio.read( file ); for( int i = 0; i < sound.length; i++ ) sound[i] = 0.5*sound[i]; StdAudio.play(sound); 27

28.  How would we double the volume of a sound in code?  Be careful!  Remember that the max is 1.0 and the min is -1.0  If you go outside of that range, you should limit the values double[] sound = StdAudio.read( file ); for( int i = 0; i < sound.length; i++ ) { sound[i] = 2*sound[i]; if( sound[i] > 1.0 ) sound[i] = 1.0; else if( sound[i] < -1.0 ) sound[i] = -1.0; } StdAudio.play(sound); 28

29.  We can take a sound:  And shrink that sound in half the time (while also doubling its frequency): 29

30.  How would we double the speed of a sound in code?  We are just taking every other sample  We are throwing out half the information double[] sound = StdAudio.read( file ); double[] speed = new double[sound.length/2]; for( int i = 0; i < speed.length; i++ ) speed[i] = sound[2*i]; StdAudio.play(speed); 30

31.  We can take a sound:  And stretch that sound to double the time (while also cutting its frequency in half): 31

32.  How would we stretch that sound to double in code?  Often times, this can sound terrible  We are doubling the length of the sound, but we have no extra information  We are just filling in holes in the samples with copies double[] sound = StdAudio.read( file ); double[] slow = new double[sound.length*2]; for( int i = 0; i < slow.length; i++ ) slow[i] = sound[i/2]; StdAudio.play(slow); 32

33.  Integers are easy  You can apply the same ideas to non-integer speed-ups and slow-downs  Smarter algorithms, like the ones used by professionals, may do some averaging or other tricks to retain sound quality 33

34.  To add noise to a waveform, add small random numbers to the wave  Make sure that the random numbers are both positive and negative (with a mean of 0)  You can turn this wave:  Into this wave: 34

35.  How can we turn this wave…  Into this wave?  That is much harder! 35