1
Q
Why do we need to locate the sources of sounds?
A
- The location of sound may be important information in itself
- The location of a sound may orient visual attention
- Sound location can separate sequences of sounds arising from diff locations
- Hearing is an anchor into the attention system - useful for survival
2
Q
Objectives
A
- Understand the cues that listeners use to localise sources of sound in the horizontal plane - inter-aural timing diffs, inter-aural level diffs
- Understanding how listeners can distinguish sounds coming from the front and back
- Understanding how listeners can distinguish sounds coming from above and below
- Understand the principle of the precedence effect and how it contributes to localisation in rooms
3
Q
The coordinates of spatial hearing: azimuth and elevation
A
Two planes
- Horizontal calculate diffs in azimuth (calculate diffs from left to right)
- Median plane: elevation is calculated from 90 to -90
4
Q
Minimum audible angle
A
- Present a pure tone at 0 degrees azimuth
- Present 2nd tone right after 1st and it is presented on the left then the right
- Decision to say on which side was the 2nd tone presented
- Manipulated until you can no longer tell if the 2nd tone is coming from the left or the right
5
Q
MAA
A
- MAA is 1 degree for tones coming from straight ahead at low frequencies
- MAA is larger (worse) around 1500 Hz, improves at higher frequencies but it not as small as it is at low frequencies
- MAA is larger (worse) for sounds coming from non-frontal locations
6
Q
How do we know where the sound is coming from?
A
- Know where it is coming from due to speed of sound - 340 m/s so technically there is a diff between time arrival in left ear compared to right ear
- When the source of sound is on the left, the sound reaches the left ear before the right ear- this is the Interaural Timing Difference (ITD)
- When the source of sound is on the left, the head shields the right ear from the sound- this is the ‘head shadow’, it generates an Interaural Level Difference (ILD)
7
Q
How does ITD vary with azimuth?
A
- When a sound comes from one side of an adult head (azimuth +90 degrees), the ITD is 0.65 milliseconds
- That duration will vary with the width of the head
- Animals for example will have a diff resolution with smaller heads
8
Q
What is the smallest ITD that listeners an distinguish from an ITD of 0 microseconds?
A
- 10 …
- Corresponds to a diff of 1 degree from straight ahead
- The auditory system has exquisite sensitivity to be inter-aural timing diffs
- 1st tone presented with ITD of 0
- 2nd tone has a non-0 ITD with left ear or right ear leading
- Task: has the 2nd tone moved to the left or right
- Manipulate ITD until you can no longer tell if it’s coming from the left or right ear
- Can detect very small diffs
9
Q
How does ILD vary with azimuth?
A
- When a sound comes from one side of an adult head (azimuth + 90 degrees), the ILD is about 20 dB at 6000 Hz, diminishing to 0 dB at 200 Hz
- What are the level diffs for sound coming from diff locations?
- Graphs show direction of sound and interaural intensity level
- The ILD will vary enormously as a function of the frequency of the tone
- High frequency tone means healthy ILD
- ILDs exist for high frequency sounds but not low frequency
- Level will vary with the head size but not as much as ITDs
10
Q
Obstacle of intensity
A
- Head is an obstacle to intensity of sound
- High frequency sound will bounce off
- High frequencies: the head interrupts the propagation of the sound waves -> high ILDs
- Low frequencies: the head doesn’t interrupt the propagation of the sound wave as much -> low ILDs
11
Q
A
12
Q
What is the smallest ILD that listeners can distinguish from an ILD of 0dB
A
- 1dB
- Sensitivity is good across a wide range of frequencies
- Despite being materially irrelevant for low frequencies
- Getting rid of head shadow problem
- 1st tone presented as 0 ILD
- 2nd tone has a non 0 ILD
- Manipulate ILD and decide whether 2nd tone has moved to right or left - what is smallest diff you can hear the diff from = 1ILD
13
Q
ITDs and ILDs…
A
- Are there for us to use and we use them as cues to localisation, although ILDs are only usable at high frequencies
- ITDs are restricted to low frequencies
- Can see that ITDs and ILDs are complementary
14
Q
ITDs are unambiguous at what frequencies
A
Low frequencies
15
Q
When are ITDs ambiguous?
A
- About 770 Hz
- 1300 microseconds is exactly double the amount of time it takes for sound to reach left ear rather than right ear
- ITDs are high frequency specific
- Principle: pairing of similar co-occurring patterns
16
Q
ITDs are misleading above 770 Hz
A
- Increase the frequency more, here all of a sudden the 2 waves look like they’re paired up differently
- Pairing the wrong peaks within this analysis
17
Q
ITDs are less ambiguous for…
A
- ITDs are less ambiguous for modulated tones (complex tones)
- At high frequency it will have the typical misperceived pairing
- Similar looking patterns…
- Problem of high frequencies ITDs will disappear
18
Q
Interim summary
A
- To localise pure tones in azimuth
- Use ITDs at low frequencies (<750 Hz)
- Use ILDs at higher frequencies (>1500 Hz)
- To localise frequency-modulated (complex) tones:
- ITDs can be used over a wider frequency range
- Most natural sounds are frequency-modulated, ITDs are the dominant cue for localising them
19
Q
How are ITDs calculated by the brain?
A
- The auditory nervous system
- The superior olive analyses the location of sources of sound.
- This happens early in the ascending auditory system because it relies on very precise timing- of the order of millionths of a second
20
Q
Measuring ITDs: delay lines and coincidence detectors
A
- For each band of frequencies we have an array of delay lines and coincidence detectors
- These are neurons probably located in the superior olives
- Action potential will run through the delay lines
- Connecting delay lines are neural coincidence detectors
- Only fire if activated by the 2 delay lines at the same time
- As the sound continues to travel it will activate the neural coincidence detectors at the same time
21
Q
Sound from in front
A
Only fires if perceived from azimuth 0
22
Q
Sound from 30 degrees to the right
A
- Might be specialising in azimuth +30
- All delay lines specialise for a specific location in space
23
Q
Delay lines and coincidence detectors
A
- Probably how localisation in azimuth is implemented in barn owls
- Terrestrial mammals, including humans, are likely to use an ‘opponent-process’ system
24
Q
Opponent-process analysis
A
- The idea that there are just 2 channels
- 1 tuned to the left half of the auditory space, the other tuned to the right half of auditory space
- The system calculates the diff between the responses of the 2 channels to work out where a sound is coming from (in the frontal horizontal plane)
- There are 2 subsystems as opposed to many frequency specific channels
- One channel cued to the left side of space
- The other system cued to right half of auditory space so responding maximally to +ve azimuth
25
Delay lines and coincidence detectors cont.
- Large number of detectors, each tuned to a specific and narrow ITD
- ITD determined by which detector fires
- Found in barn owls and some birds
26
Opponent process analysis cont.
- 2 populations of detectors broadly tuned to -ve vs +ve ITDs
- ITD is determined by how much the detectors within a channel fire relative to the detectors in the other channel
- Found in mammals (including humans)
27
Front-back ambiguity
- Sounds coming from directly in front or directly behind generate an ITD of 0
- ITDs are useless for saying whether it's front or back
- Listeners disambiguate the location of such sounds in pinna (if familiar sound) + also by rotating the head -> if you turn the head clockwise and the ITD favours the left ear then the sound is coming from the front
28
Elevation
- ITDs and ILDs are unhelpful for elevation diffs
- Because a sound reaches both ears at the same time
29
A possible solution of elevation
- Reflections of sound within the pinna boost energy at some frequencies and reduce energy at other frequencies, depending on the elevation of the sound source.
- The shape of the pinna also produces unique resonance patterns.
- That helps listeners interpret elevation, especially if they are familiar with the sound
- Attenuate and amplify diff frequencies before they enter the canal
- If listeners use those cues, then it should be possible to disrupt their judgements of elevation by modifying their pinnae
- Implants which will modify some of the faults
30
Judgements of elevation with modified pinnae
- 4 ppts tested on their perception of azimuth and elevation over several weeks, with their own pinnae, artificial pinnae and their own pinnae again
- Judgements of elevation are severely disrupted by modifying a listener's pinnae
- But accuracy improves with experience
- There is no 'after-effect' accuracy returns to normal as soon as the moulds are removed
31
Elevation summary
- The frequency spectrum of an incoming sound is modulated by the pinna.
- That modulation varies systematically with elevation
- Therefore, listeners can use pinna-specific modulations of frequency spectra to determine the elevation of sources of sound
32
VR and virtual acoustic space
- ILDs can create a fairly compelling auditory space, but they are limited to the horizontal plane
- ILDs and ITDs provide more precision, but they are also limited to the horizontal plane
33
How do we simulate elevation in VR?
- Need to use a head-related transfer functions, which are a like-for-like reproduction of the binaural signal taking into account head-related resonance patterns
- Microphones in ears
- Record sounds from manikin
- Record sounds from multiple locations from 2 ears
34
VR and VAS continued
- Manikin: HRTF take everything into account: the head shadow and the shape of the pinna
- The binaural recordings from the manikin are played back over headphones to create an externalised VAS
- Main limitation: HRTF are usually based on the manikin's pinnae
- Rotate crescent and collect info on what is perceived
- Use an echoic chamber
- If you record from manikin and playback to person you will be able to tell elevation due to brute force of all the cues
35
VR and VAS: HRTF extraction
- Binaural recording of sounds coming from various spatial locations through manikin pinnae vs own pinnae
- Testing: audio (generic or individual) played back through headphones, performance with headphones compared with performance with loudspeakers (ground truth)
36
VR and VAS graphs
- Generic: manikin HRTFs (headphones)
- Individual: ppt-specific HRTFs (headphones)
- Ground truth: loudspeakers
- % FB: front-back reversal errors
- Individual HRTF still not as good as ground truth- missing room reverberation, true sound resolution
- More mistakes made with generic pinna rather than own pinna
- Even with own pinna info, you will still make some mistakes
37
An artist cuts off both pinnae- what is the result?
They have difficulty working out whether sounds are coming from above or below
38
Localisation in rooms
- Direct sound: what ears are picking up from voice as it comes straight to you on a straight line
- Reverberant sound: a copy of the direct sound multiplied lots of times as a function of the shape of the environment
- Have to decide which copy you should trust in terms of sound localisation as all copies have diff ITDs and ILDs- which you should trust and why
39
Echo suppression: the precedence effect
- Simulate a direct sound coming from the left, followed by its echo (reverberant sound) coming from the right a few milliseconds later
- 1st click left suggests click is coming from left
- 2nd click right suggests click is coming from the right
- 2nd click is cancelled because it is perceived as an echo of the 1st click- this is only true if the delay between 2 sounds is <30 mw
40
What can the precedence effect be exploited to do
Reduce masking from a competing voice
41
Release from masking through the precedence effect
- Try to understand female voice on left
- Male voice on right heard first will be direct sound and the 2nd male voice will be treated as an echo and therefore suppressed
- A sound perceived as an echo is suppressed- precedence effect
- The precedence effects is most effective when the delay is between the direct sound and the reverberant sound is less than 30ms
- The precedence effect can be used to attenuate masking
P+C 2: Hearing
flashcards
Decks in class (4)
# Cards
