EE E6820: Speech & Audio Processing & Recognition Lecture 2: - - PowerPoint PPT Presentation

ee e6820 speech audio processing recognition lecture 2
SMART_READER_LITE
LIVE PREVIEW

EE E6820: Speech & Audio Processing & Recognition Lecture 2: - - PowerPoint PPT Presentation

EE E6820: Speech & Audio Processing & Recognition Lecture 2: Acoustics 1 The wave equation 2 Acoustic tubes: reflections & resonance 3 Oscillations & musical acoustics 4 Spherical waves & room acoustics Dan Ellis


slide-1
SLIDE 1

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 1

EE E6820: Speech & Audio Processing & Recognition

Lecture 2: Acoustics

The wave equation Acoustic tubes: reflections & resonance Oscillations & musical acoustics Spherical waves & room acoustics

Dan Ellis <dpwe@ee.columbia.edu> http://www.ee.columbia.edu/~dpwe/e6820/ Columbia University Dept. of Electrical Engineering Spring 2002

1 2 3 4

slide-2
SLIDE 2

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 2

Acoustics & sound

  • Acoustics is the study of physical waves
  • (Acoustic) waves transmit energy without

permanently displacing matter (e.g. ocean waves)

  • Same math recurs in many domains
  • Intuition: pulse going down a rope

1

slide-3
SLIDE 3

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 3

The wave equation

  • Consider a small section of the rope:
  • displacement is y(x), tension S, mass

ε

·dx → lateral force is

φ1 φ2 S S x y ε

Fy S φ2 ( ) sin ⋅ S φ1 ( ) sin ⋅ – = S = x2

2

∂ ∂ y x d ⋅ ⋅

slide-4
SLIDE 4

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 4

Wave equation (2)

  • Newton’s law:
  • Call

(tension to mass-per-length) hence the wave equation : .. partial DE relating curvature and acceleration

F ma = S x2

2

∂ ∂ y x d ⋅ ⋅ ε x d t2

2

∂ ∂ y ⋅ = c2 S ε ⁄ = c2 x2

2

∂ ∂ y ⋅ t2

2

∂ ∂ y =

slide-5
SLIDE 5

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 5

Solution to the wave equation

  • If

then also works for Hence, general solution:

y x t , ( ) f x ct – ( ) = x ∂ ∂y f' x ct – ( ) = x2

2

∂ ∂ y f'' x ct – ( ) = t ∂ ∂y c – f' x ct – ( ) ⋅ = t2

2

∂ ∂ y c2 f'' x ct – ( ) ⋅ = y x t , ( ) f x ct + ( ) = c2 x2

2

∂ ∂ y ⋅ t2

2

∂ ∂ y = y x t , ( ) y+ x ct – ( ) y– x ct + ( ) + = ⇒

slide-6
SLIDE 6

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 6

Solution to the wave equation (2)

  • and

are travelling waves

  • shape stays constant but changes position:
  • c is travelling wave velocity ( ∆x / ∆t )
  • y+ moves right, y– moves left
  • resultant y(x) is sum of the two waves

y+ x ct – ( ) y– x ct + ( )

time 0: x y y+ y- time T: y+ y- ∆x = c·T

slide-7
SLIDE 7

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 7

Wave equation solutions (3)

  • What is the form of y+, y– ?
  • any doubly-differentiable function will satisfy

wave equation

  • Actual waveshapes dictated by

boundary conditions

  • y(x) at t = 0
  • constraints on y at particular x’s

e.g. input motion y(0, t) = m(t) rigid termination y(L, t) = 0

x y

y(0,t) = m(t) y+(x,t) y(L,t) = 0

slide-8
SLIDE 8

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 8

Terminations and reflections

  • System constraints:
  • initial y(x, 0) = 0 (flat rope)
  • input y(0, t) = m(t) (at agent’s hand) (→ y+)
  • termination y(L, t) = 0 (fixed end)
  • wave equation y(x,t) = y+(x - ct) + y–(x + ct)
  • At termination:

y(L, t) = 0 → y+(L - ct) = – y–(L + ct) i.e. y+ and y– are mirrored in time and amplitude around x=L →inverted reflection at termination →simulation [travel1.m]

x = L y(x,t) = y+ + y– y+ y–

slide-9
SLIDE 9

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 9

Acoustic tubes

  • Sound waves travel down acoustic tubes:
  • 1-dimensional; very similar to strings
  • Common situation:
  • wind instrument bores
  • ear canal
  • vocal tract

2

x pressure

slide-10
SLIDE 10

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 10

Pressure and velocity

  • Consider air particle displacement

:

  • Particle velocity

hence volume velocity

  • Air pressure

ξ x t , ( )

x

ξ(x) v x t , ( ) t ∂ ∂ξ = u x t , ( ) A v x t , ( ) ⋅ = p x t , ( ) 1 κ

x ∂ ∂ξ ⋅ =

slide-11
SLIDE 11

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 11

Wave equation for a tube

  • Consider elemental volume:
  • Newton’s law:
  • Hence

x

Area dA Force p·dA Force (p+∂p/∂x·dx)·dA Volume dA·dx Mass ρ·dA·dx

F ma = x ∂ ∂p – dx dA ⋅ ⋅ ρdAdx t ∂ ∂v ⋅ = x ∂ ∂p ⇒ ρ t ∂ ∂v – = c2 x2

2

∂ ∂ ξ ⋅ t2

2

∂ ∂ ξ = c 1 ρκ

  • =
slide-12
SLIDE 12

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 12

Acoustic tube traveling waves

  • Traveling waves in particle displacement:
  • Call
  • Then pressure:
  • Volume velocity:
  • (Scaled) sum and difference of traveling waves

ξ x t , ( ) ξ+ x ct – ( ) ξ- x ct + ( ) + = u+ α ( ) cA α ∂ ∂ ξ+ α ( ) – = Z0 ρc A

  • =

p x t , ( ) 1 κ

x ∂ ∂ξ ⋅ Z0 u+ x ct – ( ) u- x ct + ( ) + [ ] ⋅ = = u x t , ( ) A t ∂ ∂ξ ⋅ u+ x ct – ( ) u- x ct + ( ) – = =

slide-13
SLIDE 13

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 13

Acoustic tube traveling waves (2)

  • Different residuals for pressure and vol. veloc.:

x u+ u- u(x,t) = u+ - u- p(x,t) = Z0[u+ + u-] c c Acoustic tube

slide-14
SLIDE 14

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 14

Terminations in tubes

  • Equivalent of ‘fixed point’ for tubes?
  • Open end is like fixed point for rope:

reflects wave back inverted

  • Unlike fixed point, solid wall reflects traveling

wave without inversion

u0(t)

(Volume velocity input)

Solid wall forces u(x,t) = 0 hence u+ = u- hence u+ = -u- Open end forces p(x,t) = 0

slide-15
SLIDE 15

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 15

Standing waves

  • Consider (complex) sinusoidal input:
  • At any point, values will have form
  • Hence traveling waves:

where (spatial frequency, rad/m) (wavelength )

  • Pressure / vol. veloc. resultants show

stationary pattern: standing waves

  • even when |A| ≠ |B|

→simulation [sintwavemov.m]

u0 t ( ) U0 e jωt ⋅ = Ke j ωt

φ + ( )

u+ x ct – ( ) A e

j kx – ωt φA + + ( )

= u- x ct + ( ) B e

j kx ωt φB + + ( )

= k ω c ⁄ = λ c f ⁄ 2πc ω ⁄ = =

slide-16
SLIDE 16

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 16

Standing waves (2)

  • For lossless termination (|u+| = |u-|),

have true nodes & antinodes

  • Pressure and vol. veloc. are phase shifted
  • in space and in time

U0 ejωt

kx = π x = λ / 2 pressure = 0 (node) vol.veloc. = max (antinode)

slide-17
SLIDE 17

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 17

Transfer function

  • Consider tube excited by

:

  • sinusoidal traveling waves must satisfy

termination ‘boundary conditions’

  • satisfied by complex constants A and B in
  • standing wave pattern will scale with input

magnitude

  • point of excitation makes a big difference

u0 t ( ) U0 e jωt ⋅ = u x t , ( ) u+ x ct – ( ) u- x ct + ( ) + = Ae j

kx – ωt + ( )

Be j kx

ωt + ( )

+ = e jωt Ae jkx

Be jkx + ( ) ⋅ =

slide-18
SLIDE 18

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 18

Transfer function (2)

  • For open-ended tube of length L excited at x = 0

by : (works at x = 0)

  • i.e. standing wave pattern

e.g. varying L for a given ω (and hence k):

U0e jωt u x t , ( ) k L x – ( ) cos kL cos

  • U0e jωt

⋅ = k ω c

  • =

    U0 ejωt U0 ejωt

U0 U0 UL UL

slide-19
SLIDE 19

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 19

Transfer function (3)

  • Varying ω for a given L:
  • at x = L,
  • Output vol. veloc. always larger than input
  • Unbounded for

i.e. resonance

u L t , ( ) u 0 t , ( )

  • 1

kL cos

  • 1

ωL c ⁄ ( ) cos

  • =

=

L u(L) u(0) u(L) u(0)

at ωL/c = (2r+1)π/2, r = 0,1,2...

L 2r 1 + ( ) πc 2ω

  • 2r

1 + ( )λ 4

  • =

=

slide-20
SLIDE 20

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 20

Resonant modes

  • For lossless tube

with , m odd, λ wavelength, is unbounded, meaning:

  • transfer function has pole on frequency axis
  • energy at that frequency sustains indefinitely
  • e.g 17.5 cm vocal tract, c = 350 m/s

→ ω0 = 2π · 500 Hz (then 1500, 2500 ...)

L m λ 4

= u L ( ) u 0 ( )

  • L = λ0/4

L = 3 · λ1/4 → ω1 = 3ω0

slide-21
SLIDE 21

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 21

Scattering junctions

  • Solve e.g. for u-

k and u+ k+1: (generalized term.)

Area Ak Area Ak+1 At abrupt change in area:

  • pressure must be continuous

pk(x, t) = pk+1(x, t)

  • vol. veloc. must be continuous

uk(x, t) = uk+1(x, t)

  • traveling waves

u+k, u-k, u+k+1, u-k+1 will be different u+k u-k u+k+1 u-k+1

+ +

u+k u-k u+k+1 u-k+1 2r 1 + r 1 - r 1 + r 2 r + 1 r - 1 r + 1 r = Ak+1 Ak “Area ratio”

slide-22
SLIDE 22

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 22

Concatenated tube model

  • Vocal tract acts as a waveguide
  • Discrete approx. as varying-diameter tube:

Glottis u0(t) Glottis Lips uL(t) Lips x = 0 x = L

x

Ak, Lk Ak+1, Lk+1

slide-23
SLIDE 23

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 23

Concatenated tube resonances

  • Concatenated tubes → scattering junctions

→ lattice filter

  • Can solve for transfer function - all-pole
  • Approximate vowel synthesis from resonances

sound example? ah ee oo

+ +

u+k u-k

e-jωτ1 e-jωτ1

+ +

e-jωτ2 e-jωτ2

+ + + +

e-jω2τ1

+ + + +

e-jω2τ2

  • 1
  • 0.5

0.5 1

  • 1
  • 0.5

0.5 1 Imaginary part

slide-24
SLIDE 24

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 24

Oscillations & musical acoustics

  • Pitch (periodicity) is essence of music:
  • why? why music?
  • Different kinds of oscillators:
  • simple harmonic motion (tuning fork)
  • relaxation oscillator (voice)
  • string traveling wave (plucked/struck/bowed)
  • air column (nonlinear energy element)

3

slide-25
SLIDE 25

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 25

Simple harmonic motion

  • Basic mechanical oscillation:
  • Spring + mass ( + damper)
  • e.g. tuning fork
  • Not great for music:
  • fundamental (cosωt) only
  • relatively low energy

x ˙˙ ω2x – = x A ωt ϕ + ( ) cos = m x F = kx ζ ω2 = k m

slide-26
SLIDE 26

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 26

Relaxation oscillator

  • Multi-state process:
  • one state builds up potential (e.g. pressure)
  • switch to second (release) state
  • revert to first state etc.
  • e.g. vocal folds:

( http://www.medicine.uiowa.edu/otolaryngology/cases/normal/normal2.htm )

  • Oscillation period depends on force (tension)
  • easy to change
  • hard to keep stable

→less used in music

p u

slide-27
SLIDE 27

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 27

Ringing string

  • e.g. our original ‘rope’ example
  • Many musical instruments
  • guitar (plucked)
  • piano (struck)
  • violin (bowed)
  • Control period (pitch):
  • change length (fretting)
  • change tension (tuning piano)
  • change mass (piano strings)
  • Influence of excitation ... [pluck1a.m]

L tension S mass/length ε ω2 = π2 S L2 ε

slide-28
SLIDE 28

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 28

Wind tube

  • Resonant tube + energy input
  • e.g. clarinet
  • lip pressure keeps reed closed
  • reflected pressure wave opens reed
  • reinforced pressure wave passes through
  • Finger holes determine first reflection

→ effective waveguide length e.g. energy nonlinear element scattering junction (tonehole) (quarter wavelength) acoustic waveguide

ω = π c 2 L

slide-29
SLIDE 29

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 29

Room acoustics

  • Sound in free air expands spherically:
  • Spherical wave equation:

solved by

  • Intensity

falls as

4

radius r

r2

2

∂ ∂ p 2 r

  • r

∂ ∂p ⋅ + 1 c2

  • t2

2

∂ ∂ p ⋅ = p r t , ( ) P0 r

  • e j ωt

kr – ( )

⋅ = p2 ∝ 1 r2

slide-30
SLIDE 30

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 30

Effect of rooms (1): Images

  • Ideal reflections are like multiple sources:
  • ‘Early echos’ in room impulse response:
  • Actual reflection may be href(t), not δ(t)

source listener virtual (image) sources reflected path t hroom(t) direct path early echos

slide-31
SLIDE 31

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 31

Effect of rooms (2): modes

  • Regularly-spaced echos behave like acoustic

tubes:

  • Real rooms have lots of modes!
  • dense, sustained echos in impulse response
  • complex pattern of peaks in frequency response
slide-32
SLIDE 32

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 32

Reverberation

  • Exponential decay of reflections:
  • Frequency-dependent
  • greater absorption at high frequencies

→ faster decay

  • Size-dependent
  • larger rooms → longer delays → slower decay
  • Sabine’s equation:
  • Time constant as size, absorption [e.g.]

t hroom(t) ~e-t/T

RT60 0.049V Sα

  • =
slide-33
SLIDE 33

E6820 SAPR - Dan Ellis L02 - Acoustics - 2002-02-04 - 33

Summary

  • Travelling waves
  • Acoustic tubes & resonance
  • Musical acoustics & periodicity
  • Room acoustics & reverberation