COURSE NOTES 7
Localization, Reverberation and Speech Synthesis

Localization

• Multiple output channels placed in various locations in a listening space (stereophonic, quadrophonic, surround sound)

• Varying amplitudes and phase of sounds through each output channel gives the illusion of movement of sound

Localization Cues

• Azimuth angle (direction on the horizontal plane from the listener)

• Elevation (direction on the vertical plane from the listener)

• Distance (from the listener)

Localization through Varying Intensity

• The intensity of a sound is proportional to the square of its amplitude

• The intensity of a sound is proportional to the inverse of the square of the distance of the sound from the listener

Example

• Let the amplitude of two speakers be G₁ (right) and G₂ (left).

• In order to maintain a sound with equal intensity c distributed between the two speakers, then G₁² + G₂² = c

• Make G₂ proportional to cos A and G₁ proportional to sin A, where A is the angle from the listener to the virtual sound source along the horizontal plane.

Localization in Csound

instr 1
k1          linseg 0, p3, 1
a1          oscil p4, p5, 2
a2,a3,a4,a5 pan a1, k1, k1 3
            outq a2, a3, a4, a5
endin

f2 0 2048 10 15 0 5 0 3
f3 0 8193 -8 0 819 .122 819 .294 819 .517 819 .769 819 1 819 1.154 819 1.207 819 1.176 819 1.098 822 1

Other Localization Techniques

• Interaural timing differences (ITD)

• Doppler effect

• Echo effects

Reverberation

• The process of dispersion of sound due to the multiple reflections of sound within the musical enclosure

• The properties of the enclosure determine the effect of the reverberation

• Example: carpeted walls vs. tiled walls

Echo (delay) in Csound

instr 1
k1 linseg 0, p3*0.1, p4, p3*0.9, 0
a1 oscil k1, p5, 1
a2 delay a1, p3*0.5
   outs a1, a2
endin

Comb Filter

• y[n] = x[n-D] + gy[n-D], where D is the sample delay.

• The impluse response h[n] of this filter is given by h[n] = u[n-D] + gu[n-2D] + g²u[n-3D] + ...
  where u[n] is the sequence {1,0,0,0,0....}.

• The name comes from the frequency response graph, which looks like comb teeth.

• The time (sec) for output to decrease 60 dB is given by T₆₀ = 60D/(-20R log₁₀ g).

Example

• g = 0.7

• ==> T₆₀ = 20t, where t is the loop time (sec) = D/R

• If T₆₀ = 2 sec, then t = 0.1 (this means you have an echo density of 10 per second).

AllPass Filter

• y[n] = -gx[n] + (1 - g²)(x[n-D] + gy[n-D])

• The frequency response of the allpass filter is 1.

• Allpass filters have non-uniform phase response.

• Natural reverberation effects can be achieved by combining comb and allpass filters.

Reverberation in Csound

instr 1
; p4 = amp, p5 = pitch, p6 = reverb time in seconds
k1 linseg 0, p3*.1, p4, p3*.9, 0
a1 oscil k1, p5, 1
a2 comb a1, p6, .0297	; .0297 is loop time
a3 comb a1, p6, .0371
a4 comb a1, p6, .0411
a5 comb a1, p6, .0437
a6 = a2 + a3 + a4 + a5
a7 alpass a6, .09683, .005
a8 alpass a7, .03292, .0017
a9 reverb a1, p6	; reverb is same as comb/alpass combination
   outs a8, a9
endin

Speech Synthesis

• Speech consists of streams of phonemes (see handout)

• Air (noise) is expelled from the lungs and passes through the voice box (larynx) which acts like a filter

• Other parts of the vocal and nasal tract act like filters to further modify and shape the output sound

Types of Excitation Sources

• Voiced - due to the periodic vibration of the glottis (has a detectable pitch)

• Fricative (unvoiced) - due to turbulence created by constriction at some point along the vocal tract

• Plosive (unvoiced) - due to the sudden release of pressure built up behind a closure in the vocal tract

Voiced Speech

• Start with pink noise (buzz).

• Filter a locations of formants (frequencies that are emphasized during specific speech sound - see handout)

• Pass through additional filters to simulate nasalization and other vocal effects (vibrato, tremelo)

Unvoiced Speech

• Start with white noise (rand).

• Filter to simulate fricative and plosive effects due to the lips, tongue and teeth

• Much more difficult to simulate unvoiced speech by computer than voiced sounds like vowels or vowel combinations.

Speech Synthesis Example in Csound

; speech synthesis of /A/ (as in "hot")
; p4 = note amp
instr 1
iformant1 = 730
iformant2 = 1090
iformant3 = 2440
k1 linseg 0, .1, p4, p3-.2, p4, .1, 0
a1 buzz k1, 10, (sr/2)/10, 1
a2 reson a1, iformant1, 0.05*iformant1
a3 reson 0.5*a1, iformant2, 0.04*iformant2
a4 reson 0.33*a1, iformant3, 0.03*iformant3
   out a2+a3+a4
endin

f1 0 8192 10 1

COURSE INFORMATION | HOMEWORK ASSIGNMENTS
COURSE PROJECT | CS240 HELP DESK