Speech-in-Noise & Directional Mics Directional microphone technology was invented by the military during World War II. It’s concept was refined and miniaturized into hearing instruments in the late 1960’s.
May 25, 2015
Speech-in-Noise & Directional Mics
Directional microphone technology was invented by the military during World War II.
It’s concept was refined and miniaturized into hearing instruments in the late 1960’s.
Speech-in-Noise & Directional Mics
Hearing instruments designed with this technology would, theoretically, provide a better listening experience for the hearing impaired. However, the early microphones produced so much of their own noise when operating that they were of limited utility.
Speech-in-Noise & Directional Mics
As HI dispensing professionals, there are two things we must do for the hearing impaired patient/client:
1.Improve audibility2.Improve signal-to-noise ratio
Speech-in-Noise & Directional Mics
The amplified gain of hearing instruments is used to provide the audibility of sound for the hearing impaired.
However, amplified audibility does not always provide a greater opportunity for speech recognition in difficult listening environments i.e. speech-in-noise ability.
Hearing Instruments MakeSoft Compromised Sound
Into LoudCompromised Sound
Speech-in-Noise & Directional Mics
Directional microphones have proven to provide a greater opportunity of speech recognition in noisy environments.
In fact, refinements in the 1990’s provided for the remote control of the directional microphone ability by the patient/client.
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONESThey do not increase the sound input
from the front of the listener—they reduce the input of the sound behind them.
It is hoped that the desired input signal is in front of the patient/client and not behind them!
Speech-in-Noise & Directional Mics
Conventional• 1 mic, 2 ports• on BTEs only • often no “on/off”
Newer• 2 omni mics• found on ITEs• routinely “on/off”
In Any MicrophoneSound Moves Diaphragm
Diaphram
Source
Diaphram
Source Source
Sounds Hitting Both Sides
Cancel Each Other Out
Directional Microphone Function When Sounds Come From Front…
)
)
)
The Diaphragm moves
)
)
Filter
RearFront Direction of incoming sound
D )))))
From the Front
D )))))
From the Front
Diaphragm moves
D )
From the Front
D )
From the Front
Sound From BehindSound is cancelled by hitting both sides of
diaphragm
Front Rear
Acoustic Time Delay Network
Directional Microphone Function
When Sounds Come From Rear…
)
The Diaphragm cannot move
)
RearFront
((
((
(
(((
Filter slows sound
Direction of incoming sound
From Behind
(((((
• (((((
(
From Behind
• (((((
(
From Behind
• (((((
(
From Behind
• ((((()
From Behind
• (((((
)
From Behind
Diaphragm does not move
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONES
With this fundamental front—back operation, it was found that the greater the separation between the two microphone openings (ports), the more effective the speech-in-noise became.
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONESThey do not
restore normal hearing ability—they stimulate the residual ability of the patient/client.
HEARING LOSS IS PERMANENT!
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONES
There is currently new and more expensive directional technology which involves two separate microphones within the hearing instrument i.e. “dual microphone processing”.
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONESANSI has established a standard for
measurement of their effectiveness. It is calculated as the Directivity Index (DI).
The greater the DI, the more effective it is regarding the separation of the signal-to-noise.
Speech-in-Noise & Directional Mics
POLAR PLOTSThe directional performance graphs/charts
are represented as Polar Plots. Directional performance polar plots begin
by representing a Directivity Index (DI) as zero—(no DI).
It’s graphical appearance is represented as a perfect 360 degree circle on the polar plot.
Speech-in-Noise & Directional Mics
POLAR PLOTSAn increased DI number represents
the effectiveness of directional microphone activity.
The DI number increases, as polar plots reveal null points (areas where sound is reduced in intensity).
Speech-in-Noise & Directional Mics
POLAR PLOTSThese reduced intensity areas are
represented as indentations into the perfect circle.
These indentations are referred to as the null areas of intensity.
Speech-in-Noise & Directional Mics
Some PerfectlyRoundedSymmetricalPolar Plots:
OmnidirectionalCardioidSupercardioidHypercardioid
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONESMost of today’s typical directional microphones generally represent a Directivity Index (DI) of about 5dB to 6dB.
Speech-in-Noise & Directional Mics
DIRECTIONAL MICROPHONESTheir effectiveness is, of course, influenced
by the frequencies it/they receive. In other words the directivity index (DI)
varies by the different input frequencies. In fact, many algorithms for today’s
instruments have automatic reduction of low frequency inputs when directional microphone activity is initiated.
Speech-in-Noise & Directional Mics
DI and FrequenciesThe most important frequencies for
understanding speech are: 1kHz - 4kHz (the most important of these is 2kHz).
One can simply take average DI’s of 4 polar plots to calculate the overall microphone DI for hearing instrument performance.
Speech-in-Noise & Directional Mics
ARTICULATION INDEX To determine the best performance for speech understanding, dots within the “speech banana” were created.
This was to identify the most effective input frequencies to be “processed” by directional microphones.
Speech-in-Noise & Directional Mics
ARTICULATION INDEXThere are one hundred dots, with
each dot representing a one percent contribution towards speech intelligibility.
You will notice that most of the dots are located within the higher frequencies of the “speech banana”.
Speech-in-Noise & Directional Mics
ARTICULATION INDEX
Speech-in-Noise & Directional Mics
SPEECH UNDERSTANDING IN NOISE
Normal hearing allows for an understanding of speech-in-noise ability to be achieved fifty percent of the time; this should occur when the speech signal intensity is equal to the noise signal intensity.
Speech-in-Noise & Directional Mics
SPEECH UNDERSTANDING IN NOISE
In 1997, Meade Killion’s research discovered that with every one decibel of improvement of the speech signal over the noise signal, a ten percent improvement was realized for the ability to better understand speech in noise.
Speech-in-Noise & Directional Mics
Certainly, the five to six decibel improvement in the signal-to-noise ratio exhibited by today’s directional microphones, can reflect a fifty to sixty percent expected improvement in performance for the patient/client—if they have residual hearing ability to stimulate.
• Directional microphones objectively improve speech/noise performance
• Digital noise reduction subjectively enhances comfort in noise
Solutions for speech understanding in NOISE
Speech-in-Noise & Directional Mics
Adaptive Directionality
This technology is designed for the use of twin omni-directional microphones. It can: • Automatically shift from Omni to Dmic (Depending
upon listening environment). • Automatically switch among various polar plots
(Depending on listening situation). • Shift polar plot nulls to origin of noise (Depending
on noise source direction). NOTE: It is not necessarily statistically better…But has advantages for those with poor manual dexterity and those who cannot tell when to use a feature or what feature should be used.
Speech-in-Noise & Directional Mics
Adaptive directionality created with dual microphones has not shown any further improvement in the SNR
Multiple microphone (more than two) technology can provide greater than the five to six decibels of signal-to-noise ratio improvement.
However, is it practical?
Beam forming: Making Directional Mics Better
YetMics with more than 2 ports• eg. 3 or more mics
This is Killion’s ArrayMicTM
• heart is in right place• DI’s are about 7-10dB!
Photo provided Courtesy M. Killion
Photo provided Courtesy M. Killion
1. Directional mic transmitter
2. Ear-level FM receiver
Speech-in-Noise & Directional Mics
DIGITAL NOISE REDUCTION
•Spectral subtraction
•Phase cancellation
•Spectral enhancement
•Speech synthesis
Noise reduction: Implementation fraught with flaws
The problem: Speech & noise are mixed together
3600
2400
1200
00 300 0 300 0 300
Time in msec
Hz
/ba/ /da/ /ga/
Critical Speech Cues on Spectrogram
Speech in Quiet
Speech in Noise
Spectral Subtraction
1. The spectrum of speech & noise together is received2. During pauses in conversation, the spectrum of noise is
estimated. 3. The spectrum of speech & noise is subtracted by the
spectrum of noise only.4. Theoretically, this leaves just the spectrum of speech. Problem:
• The wide noise spectrum intersects with the speech spectrum.
• Thus, removing noise removes some of speech.
50
75
Frequency (kHz).1 1. 10
1. Speech Plus Noise Spectrum…Speech with its 6db/octave roll-offNarrow bands of noise
dB
SP
L
50
75
Frequency (kHz).1 1. 10
2. Noise Spectrum b/w Pauses of Speech…d
B S
PL
50
75
Frequency (kHz).1 1. 10
dB
SP
L3. Subtract Noise Measured During Speech Pauses
From Speech + Noise SpectrumLeaves This Speech Spectrum…
50
75
Frequency (kHz).1 1. 10
dB
SP
LThis Speech Spectrum…
Really Isn’t Too Badly Altered from Original
50
75
Frequency (kHz).1 1. 10
Problem is, Noise Spectrum is Often Wide…Intersects with Wide Speech Spectrum…
So, this combined Speech + Noise Spectrum…
dB
SP
L
50
75
Frequency (kHz).1 1. 10
Minus the Wide Noise Spectrum…
Measured during pauses in speech
dB
SP
L
50
75
Frequency (kHz).1 1. 10
dB
SP
LSubtract Noise Measured During Speech Pauses
From Speech + Noise SpectrumLeaves Only This Speech Spectrum…
50
75
Frequency (kHz).1 1. 10
Former Wide Speech Spectrum Now Badly Altered…
dB
SP
L
Phase Cancellation1. Exact time waveform of noise is measured.
2. Inverted noise phase added to the original noise waveform cancels the noise.
This phase+
Opposite phase
=
Because in the headphones:• Speech is sent directly to the eardrum from the headphone.
• Noise is sampled by the microphone outside of the headphone.
Digital hearing aids:• Do not have this luxury.• As both the speech & noise inputs are picked up by outside microphone.
Phase Cancellation used in noise reduction headphones…
Why not in hearing aids?
Why Phase Cancellation Can Work in Headphones
But Not in Hearing Aids
Speech enters directly from headphone
Noise from outsideleaks into ear canal and mixeswith speech
Noise from outsidepicked up by microphone andinverted in phase THIS CANCELS OUT THE NOISE
Spectral Enhancement
1. A digital algorithm detects spectral speech cues in noise such as vowel formants or high-frequency sibilants.
2. The algorithm deliberately enhances or amplifies these spectral speech cues. This is just a different approach from noise reduction.
The challenge for Spectral Enhancement, is the high-frequency consonants:
In noise:• The valleys b/w peaks of speech are filled
w/noise the peaks are thus less prominent• The low-frequency vowels are more intense,so these still stand out--it is easier to enhance these.
The challenge for Spectral Enhancement, is the high-frequency consonants
In noise:The valleys b/w peaks of speech are filled w/noisepeaks thus, they become less prominent.
The low-frequency vowels are more intenseSo, these still stand out. It is easier to enhance these.
Speech Synthesis
1. The digital algorithm detects spectral speech cues in noise. 2. Once a particular speech sound is detected it then adds a similar synthesized speech sound.
NOTE: It requires a “stored collection” of speech sounds. This may result into:
• A difficulty to digitally recognize speech sounds.• Overwhelming complexity. • A synthesized speech sound which can sound unnatural.
Today’s Digital Hearing Aids Use: A weak form of Spectral Subtraction is an amplitude modulation approach; this sometimes is a combo of frequency and “time-of-duration” modulation.
By subtracting the noise spectrum from the noise + speech spectrum, it may remove too much speech.
DSP algorithms have characterized waveforms. In each channel the noise has a fairly flat waveform. Over time the speech waveform fluctuates rapidly.
If noise is sensed within a channel,then gain is reduced some 5-20dB.
For Speech:Mean Intensity is Not in Middle of Range
25
50
75
Frequency (kHz).1 1K. 10
Long Term Average Speech Spectrum (LTASS)d
B S
PL
THIS IS BECAUSE SPEECH HAS ABNORMAL DISTRIBUTION OF INTENSITY
Noise Reduction Most Digital Hearing Aids Use It...
Sounds that don’t change in intensity are reduced
Sounds that do change in intensity (speech) are not reduced
How often(eg. %)
the intensityof thesound is at some
particular dB level
Speech (single talker)
Decibel Level
Noise
Speech has an odd distribution of intensity
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
Gai
n (d
B)
Noise Reduction with One Channel…
J m d b i
u
a
zv
rn onge l
ph gch
sh
k
fs
th
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
Gai
n (d
B)
Noise Reduction with One Channel…
Reduces Gain Over All Speech Hz’s!
All Speech Sounds DropJ m d b i
u
a
zv
rn onge l
ph gch
sh
k
fs
th
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
Gai
n (d
B)Noise Reduction with
Two Channels. Isn’t much better?
J m d b i
u
a
zv
rn onge l
ph gch
sh
k
fs
th
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
Gai
n (d
B)Noise reduction would reduce half of the gain
over the Speech frequencies!
Vowels Would Drop
Consonants Would NotJ m d b i
u
a
zv
rn onge l
ph gch
sh
k
fs
th
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Center Hz of Each Band
Gai
n (d
B)
Noise reduction with lots of bands/channels...
J m d b i
u
a
zv
rn onge l
ph gch
sh
k
fs
th
125 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 6500 7000
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Center Hz of Each Band
Gai
n (d
B)
Would reduce gain over smaller Hz regions
J m d
ua
zv
r
n nge l
ph
g
chsh k
f
s
th
o
b i
Digital Noise Reduction+
Directional Microphones
Noise reduction algorithms• give subjective comfort to client
Directional microphones• gives objective improvement in speech reception
Together they make a good team!• a twin-headed approach to speech-in-noise