Digital Media

Dr. Jim RowanITEC 2110

Audio

What is audio?

First, some demos

• Can you hear this?– http://www.freemosquitoringtones.org/

hearing_test/– “mosquito ring tone”

• Audio illusion “Creep”– http://www.youtube.com/watch?

v=ugriWSmRxcM

The nature of sound

First, a video from ted.comhttp://www.wimp.com/howsound/

• Three types we will discuss– 1) Environmental sound(sounds found in the environment)

• There are two special classes of audio– 2) Music– 3) Speech

The nature of sound• Environmental sounds

– Provides information about the surroundings that the human is currently in

• Music and Speech– Functionally and uniquely different than other

sounds– Music

• Carries a cultural status• Can be represented by non-sound: MIDI• Can be represented by a musical score

– Speech• Linquistic content• Lends itself to special compression

And it’s complicated…

• Converting energy to vibrations and back • Transported through some medium

– Either air or some other compressible medium• Consider speech

– Starts as an electrical signal (brain & nerves)– Ends as an electrical signal (brain & nerves)– But…

No… it’s REALLY complicated..http://en.wikipedia.org/wiki/Ear

– Starts as an electrical signal (brain & nerves) ==>– Muscle movement (vocal chords)

• Vibrates a column of air sending out a series of compression waves in the air

– Compression waves cause ear membrane to vibrate ==> – Moves 3 tiny bones ==> – Causes waves in the liquid in the inner ear ==>– Bends tiny hair cells immersed in the liquid ==> – When bent they fire ==>– Sends electrical signals to the cerebral cortex– Processed by the temporal cortex

Audio Illusions

• Play a 200 Hz pure tone– Softly at first– Gradually increase the volume– Most listeners will report that the tone drops in

pitch as the volume increases• Play a 2000 Hz pure tone

– Softly at first– Gradually increase the volume– Most listeners will report that the tone rises in

pitch as the volume increases

Why do you think…

• You can’t tell where some sounds come from (like some alarms for instance)

• You only need one sub woofer when you need at least two for everything else

• You can’t tell where sound is coming from underwater

• Two things running at the same speed make a “beating” sound

Why do you think… (cont)

• With your eyes closed you can’t tell whether a sound is in front of you or behind you

• You hear sound that isn’t there (tinnitis)• Phantom sounds

– Heard… but not there• Masking sounds

– Not simply drowning them out– Can mask a sound that occurs before the

masking sound actually starts

Why do you think… (cont)

• You can hear your name in a noisy room– Cocktail party effect– http://en.wikipedia.org/wiki/

Cocktail_party_effect– Still very much a subject of research

• http://en.wikipedia.org/wiki/Psychoacoustics

• Psychoacoustics– The study of human sound perception– The study of the psychological and

physiological affects of sound

Why? It’s complicated!

Why?It’s complicated!

• Sound is physical phenomenon that is interpreted through the human perceptual system– Wavelength affects stereo hearing

• The distance between your ears related to the wavelength

– Speed of sound affects stereo hearing• The faster the sound travels, the wider apart your ears need to

be

– You can tell where a sound comes from if • the wavelength is long enough and• the speed that sound travels is slow enough to allow the waves

arrive at your ears at different times

Processing Audio

Processing audio

• How can we look at sound?• What do you want to see?• Waveform displays

– Summed amplitude of all frequencies & time– Amplitude & frequency components at one point in

time – Amplitude & frequency & time

Summed amplitude across all frequencies & time

more examples of this form ==>

now for some other forms of audio display ==>

Amplitude & frequency components at one point in time

pipe organ audio

Amplitude & frequency & time

pipe organ audio

joe took father’s shoe bench out

Summed amplitude & time

Amplitude & frequency & time

Here… the amplitude (volume) is shown as increasingly darkening areas

Digitized audio

• As we have seen earlier this semester– Sample rate & quantization level – Reduction in sample rate is less noticeable than

reducing the quantization level• Jitter is a problem

– Slight changes in timing causes problems• 20k+ frequencies?

– Though they can’t be heard they manifest themselves as aliases when reconstructed

Audio DitheringWeird…

add noise… get better sounding result

• Add random noise to the original signal• This noise causes rapid transitioning

between the few quantized levels• Makes audio with few quantization

levels seem more acceptable

Audio processingterms to know

• Clipping– …but you don’t know how high the amplitude will be before the

performance is recorded• Noise gate

– has an amplitude threshold• Notch filter

– remove 60 cycle hum• Low pass filter• High pass filter• Time stretching (or shrinking… Limbaugh)• Pitch alteration• Envelope shaping (modifying attack)

One thing about humans…

• We can actively “filter out” what we don’t want to hear– remember the cocktail party effect?

• Over time we don’t hear the pops and snaps of a vinyl record– Have you ever recorded something that you thought

would be good only to play it back and hear the air conditioner or traffic roaring in the background?

• A piece of software can’t do this…– …not yet anyway!

Compressing sound files

• Take the opposite approach from the one you took with images– With images you can toss out the high

frequencies– With audio you can’t… high frequency

changes are highly significant

Compressing sound: Voice

• Remove silence– Similar to RLE

• Non-linear quantization• “companding”

– Quiet sounds are represented in greater detail than loud ones

• Mu-law (North America and Japan)• A-law (Europe)

– Allows a dynamic range that would require 12 bits into 8 bits

– 4096 (2**12) ==> 256 (2**8)

Compressing sound: Voice• Differential Pulse Code Modulation (DPCM)

– Related to temporal (inter-frame) video compression• It predicts what the next sample will be• It sends that difference rather than the absolute value• Not as effective for sound as it is for images

• Adaptive DCPM– Dynamically varies the sample step size

• Large differences were encoded using large steps• Small differences were encoded using small steps

Sound compressionthat is based on perception

• The idea is to remove what doesn’t matter• Based on the psycho-acoustic model

– Threshold of hearing• Remove sounds too low to be heard

– High and low frequencies not as important (for voice)

Questions?

• http://www.ted.com/talks/lang/eng/david_byrne_how_architecture_helped_music_evolve.html