Meena Ramani 04/10/06 EEL 6586 Automatic Speech Processing.

Meena Ramani

04/10/06

EEL 6586 Automatic Speech Processing

Topics to be covered

Lecture 1: The incredible sense of hearing 1The incredible sense of hearing 1

Anatomy

Perception of Sound

Lecture 2: The incredible sense of hearing 2The incredible sense of hearing 2

Psychoacoustics

Hearing aids and cochlear implants

Lecture 1:Lecture 1:The incredible sense of hearing The incredible sense of hearing

“Behind these unprepossessing flaps lie structures of such delicacy that they shame the most skillful craftsman"

-Stevens, S.S. [Professor of Psychophysics, Harvard University]

Why study hearing?

• Best example of speech recognition– Mimic human speech processing

• Hearing aids/ Cochlear implants• Speech coding

• The stapes or stirrup is the smallest bone in our body. – It is roughly the size of a grain of rice ~2.5mm

• Eardrum moves less than the diameter of a hydrogen atom – For minimum audible sounds

• Inner ear reaches its full adult size when the fetus is 20-22 weeks old.

• The ears are responsible for keeping the body in balance

• Hearing loss is the number one disability in the world. – 76.3% of people loose their hearing at age 19 and over

Interesting facts

Specifications

Frequency range: 20Hz-20kHz

Dynamic range: 0-130 dB

JND frequency: 5 cents

JND intensity: ~1dB

Size of cochlea: smaller than a dime

A

N

A

T

O

M

Y

Outer ear

Focuses sound waves (variations in pressure) into the ear canal

Pinna size:• Inverse Square Law• Larger pinna captures more of the wave • Elephants: hear low frequency sound from up to 5 miles away

Human Pinna structure: • Pointed forward & has a number of curves• Helps in sound localization• More sensitive to sounds in front

Dogs/ Cats- Movable Pinna => focus on sounds from a particular direction

Pinna /Auricle

Auditory Canal

Interaural Time Difference (ITD)

Interaural Intensity Difference (IID)

Horizontal localization

Vertical localization

Sound Localization

Outer earPinna /Auricle

Auditory Canal

Is sound on your right or left side?

Interaural differences

- The signal needs to travel further to more distant ear- More distant ear partially occluded by the head

Two types of interaural difference will emerge

- Interaural time difference (ITD)- Interaural intensity difference (IID)

Illustration of interaural differences

Leftear

Rightear

soundonset

left right

time

Leftear

Rightear

soundonset

time

arrival timedifference

Illustration of interaural differences

Leftear

Rightear

soundonset

time

ongoing timedifference

Illustration of interaural differences

Leftear

Rightear

soundonset

time

inte

nsity

diff

eren

ce

Illustration of interaural differences

Interaural time differences (ITDs)

Threshold ITD 10-20 s (~ 0.7 cm)

Interaural intensity differences (IIDs)

Threshold IID 1 dB

Thresholds

Interaural time differences (ITDs) Low frequencies

• Up to around 1500 Hz; sensitivity declines rapidly above 1000 Hz

• Smallest phase difference corresponds to the true ITD

Interaural intensity differences (IIDs) High Frequencies

• The amount of attenuation varies across frequency

• below 500 Hz, IIDs are negligible (due to diffraction)

• IIDs can reach up to 20 dB at high frequencies

D

U

P

L

E

X

T

H

E

O

R

Y

Pinna Directional Filtering

Horizontal localization

Vertical localizationSound Localization

•Pinna amplifies sound above and below differently

•Curves in structure selective amplifies certain parts of the sound spectrum

Outer earPinna /Auricle

Auditory Canal

Is sound above or below?

Pinna /Auricle

Auditory Canal

•Closed tube resonance: ¼ wave resonator

•Auditory canal length 2.7cm

•Resonance frequency ~3Khz

•Boosts energy between 2-5Khz upto 15dB

Outer ear

A

N

A

T

O

M

Y

Middle Ear

Impedance matching– Acoustic impedance of the fluid is 4000 x that of air– All but 0.1% would be reflected back

Amplification– By lever action < 3x– Area amplification [55mm2 3.2mm2] 15x

Stapedius reflex – Protection against low frequency loud sounds– Tenses muscles stiffens vibration of Ossicles– Reduces sound transmitted (20dB)

Eardrum

Ossicles

Oval window

Pressure variations are converted to mechanical motionEardrum OssiclesOval WindowOssicles: Malleus, Incus, Stapes

A

N

A

T

O

M

Y

Inner EarSemicircular Canals

Cochlea

Body's balance organsAccelerometers in 3 perpendicular planes

•Hair cells detect fluid movements•Connected to the auditory nerve

Cochlea is a snail-shell like structure 2.5 turns

3 fluid-filled parts:

•Scala tympani

•Scala Vestibuli

•Cochlear duct (Organ of Corti)

(1) Organ of Corti

(2) Scala tympani

(3) Scala vestibulli

(4) Spiral ganglion

(5) auditory nerve fibres

Semicircular Canals

Cochlea

Inner Ear

Semicircular Canals

Cochlea

Organ of CortiBasilar membraneInner hair cells and outer hair cells (16,000 -20,000)IHC:100 tiny stereocilia

The body's microphone:•Vibrations of the oval window causes the cochlear fluid to vibrate•Basilar membrane vibration produces a traveling wave•Bending of the IHC cilia produces action potentials•The outer hair cells amplify vibrations of the basilar membrane

Inner Ear

The cochlea works as a frequency analyzer It operates on the incoming sound’s frequencies

Place Theory

• Each position along the BM has a characteristic frequency for maximum vibration

• Frequency of vibration depends on the place along the BM• At the base, the BM is stiff and thin (more responsive to high Hz)• At the apex, the BM is wide and floppy (more responsive to low Hz)

32-35 mm long

4mm21mm2

Tuning curves of auditory nerve fibers

Response curve is a BPF with almost constant Q(=f0/BW)

To determine the tonotopic map on Cochlea

•Apply 50ms tone bursts every 100ms

•Increase sound level until discharge rate increases by 1 spike

•Repeat for all frequencies

Auditory Neuron

Carries impulses from both the cochlea and the semicircular canalsConnections with both auditory areas of the brainNeurons encode– Steady state sounds– Onsets or rapidly changing frequencies

Auditory Area of Brain

Auditory Neurons Adaptation

•At onset, auditory neuron fiber firing increases rapidly

•If the stimulus remains (a steady tone for eg.) the rate decreases exponentially

•Spontaneous rate: Neuron firings in the absence of stimulus

Neuron is more responsive to changes than to steady inputs

Perception of Sound

Threshold of hearing– How it is measured– Age effects

Equal Loudness curves

Bass loss problem

Critical bands

Frequency Masking

Temporal Masking

Threshold of Hearing

Hearing area is the area between the Threshold in quiet and the threshold of pain

Bekesy TrackingSTEPS:

•Play a tone

•Vary its amplitude till its audible

•Then tone’s amplitude is reduced to definitely inaudible and the frequency is slowly changed

•Continu\e

Threshold variation with age

•Presbycusis

•Hearing sensitivity decreases with age especially at High frequencies

•Threshold of pain remains the same

•Reduced dynamic range

32-35 mm long

4mm21mm2

Equal Loudness Curves

Loudness is not simply sound intensity!

Factor of ten increase in intensity for the sound to be perceived as twice as loud.

The Bass Loss Problem

Eg. Rock music

Too lowno bass

Too hightoo much bass

For very soft sounds, near the threshold of hearing, the ear strongly discriminates against low frequencies.

For mid-range sounds around 60 phons, the discrimination is not so pronounced

For very loud sounds in the neighborhood of 120 phons, the hearing response is more nearly flat.

Elephants

Sound Production

A a typical male elephant’s rumble is around an average minimum of 12 Hz, a female's rumble around 13 Hz and a calf's around 22 Hz.

Produce sounds ranging over more than 10 octaves, from 5 Hz to over 9,000 Hz

Produce very gentle, soft sounds as well as extremely powerful sounds. (112dB recorded a meter away)

Hearing

Wider tympanic membranes Longer ear canals (20 cm) Spacious middle ears.

Low frequency detection