Sound Processing Amira EL Shennawy M.D Professor of Audio vestibular medicine- Cairo University

Alleviating deafness with CI • HL is most commonly associated with degeneration or loss of cochlear hair cells, rather than a problem with neurons.

• CI directly stimulates these neurons. • Problem of converting acoustic waves to electrical impulses is solved by Signal Processing

Theories of Hearing • Place theory: Tuning of different parts of the cochlea to different frequencies. Information about the particular frequency of an incoming sound wave is coded by the place along the cochlear partition with the greatest mechanical displacement.

• Temporal coding: Tuning in which information about the particular frequency of an incoming sound wave is coded by the timing of neural firing as it relates to the period of sound. Phase locking: firing of a single neuron at one distinct point in the period (cycle) of a sound wave at a given frequency. Existence of phase locking means that the firing pattern of an AN fiber carries a temporal code ( true for low frequencies only).

• Duplicity : below 1-4 KHz → Periodicity above 1-4 KHz → Place

Normal Hearing • Intensity domain : Sound 120 dB Compression by stapedius reflex Amplification OHCs

. Spectral domain :

Electrical hearing • Intensity domain : IDR : 8 – 80 dB • Spectral domain :

20 – 20 KHz 40 band pass filters 30,000 nerve fibers

250 Hz – 8 KHz 12 – 22 channels Sequential or paired stimulation

. Temporal domain : Fine speech structure

Temporal domain : Fine speech structure up to 2.9 KHz

• • • •

Multichannel CI : Microphone Transmitter Electrodes that receive signals from transmitter to stimulate the neurons.

Just as the BM of the cochlea resolves a wave into its frequency components , so does a signal processor!

• Signal processor divides the acoustic signal into component frequencies , each transmitted to an assigned electrode. High frequencies transmitted to basal electrodes and low freq components to apical ones. • Loud sounds produce high amplitude electrical excites greater number of nerve fibers while quite ones excite less.

Compressed analog :

•

Uses continuous analog signals as stimuli.

•

Represents waveform , no specific feature of

the input. •

Signals compressed to the narrow dynamic range of electrically evoked hearing using AGC.

•

Output filtered into 4 frequency bands.

•

Simultaneous

presentation

(interaction among channels).

of

stimuli:

• To avoid uncontrolled channel interaction, pulsatile coding strategies have been developed → deliver non simultaneous pulses on multiple electrode arrays. • Was introduced based on the explicit extraction of features important for the recognition of speech.

FEATURE

EXTRACTION

• Feature extraction: Aim to reduce redundancy of information contained in speech signal by extracting most imp. cues for speech perception. • Quality (timbre ) of complex sounds is determined by spectral shape. • Vowel has a spectrum or peak at specific frequencies = formant frequency.

F0\F2 :extract and code fundamental frequency (FO) & second formant freq (F2) only. Detector estimate F0 of voiced speech sounds from output of 270 Hz low pass filter . A second detector used to estimate the second formant in the output of a band pass filter spanning the range of F2 ( 1000-4000 ).

F0\F1\F2 : F1 ( 300-1000 Hz) : F1 stimulates apical electrodes & F2 for basal electrodes.

Multipeak (MPEAK): More robust , accurate extraction of F0 Digital processing applied to: improve accuracy and resolution when extracting spectral peaks and amplitudes in the frequency ranges of 1st, 2nd formant Mini speech processor (MSP) implemented differ in :

a)

Refined technique in feature extraction

b)

Change of F1 band from (300-1000Hz) –(280-1000Hz)

c)

Change of F2 band from ( 1000-4000Hz)- (800- 4000Hz)

d)

Rates of stimulation (200-300 pps) as opposed to 100pps previously used .

e)

High frequency emphasis.

Spectral maxima sound processor strategy (SMSP): •

Microphone - AGC bank of 16 band pass filters (spanning the range from 250-5.400Hz)

•

Channels with low center frequencies deliver

output to elec at apical positions & channels with high center frequencies to elec at basal positions •

A post processor is used to scan the outputs of the channels , it selects the 6 channel outputs that have greatest amplitudes among the set for stimulation

in that cycle.

Spectral peak strategy (SPEAK): •

Uses a bank of 20 digitally programmable analog filter each analyzing the amplitude of speech within a specified frequency range.

•

Electrodes corresponding to the filters containing the greatest energy (maxima) are stimulated in tonotopic order.

•

Number of maxima and rate of stimulation are adaptive to preserve the fine spectral detail of speech . By stimulating 10 electrodes for complex signals , and increasing the rate and stimulating fewer electrodes for signals with limited spectral information. Pulse rate 500 - 1.200 pps.

Continuous Interleaved Sampling (CIS) Combines advantage of non simultaneous pulses on multiple electrodes with preservation of fine temporal structure of original analog signal by using brief pulses at high rate. Differs from SPEAK in number of bands is not fixed and is lower ( 4, 6 ,8 or 12). Information from all filters is used for stimulation not depending on energy level. Is a strategy with a very high stimulating rate ( up to 1800 Hz) but a small number of processed bands.

•

Microphone → AGC → Pre- emphasis filter : Which

Decreases Frequency component <1.2KHz: help weak consonants (predominant frequency above 1.2KHz compete with vowels which are intense and have strong component below 1.2KHz). Therefore represents voicing information in a more natural way. Used in the speech manufacturers.

processors

of

all

major

CI

Advanced Combination Encoders (ACE): •

Combines the strengths of CIS (high ++ rate) and SPEAK (high number of processed bands).

•

Uses 22 implanted electrodes.

•

Based on the N of M principle: n= specific no of spectral peaks , m= no of different band pass filters.

Stimulating strategy using virtual channel

The above strategies only use fixed channels to reproduce the original sound spectrum. There are, however, around 30,000 auditory nerve fibers in a human ear, but only 12 –22 electrodes can currently be implanted into a CI user’s ear to generate 16–22 fixed channels. Thus exciting a small number of specific auditory nerve fibers.

HiRes120 When two (or more) neighboring electrodes are stimulated in a suitable manner, intermediated channels, also known as virtual channels, are created between the electrodes. This technique uses current steering where each electrode has an independent power source to allow the current to be delivered simultaneously to more than one electrode. 8 stimulation sites between the 16 electrodes × 15 spaces between electrodes = 120.

• These virtual channels can enable CI users to perceive different frequencies between two fixed channels. • HiRes – S : stimulates each channel sequentially

• HiRes – P : stimulates 2 channels simultaneously (paired) • HiRes –optima : optimize battery life.

The Elements of Sound – Envelope and Fine Structure

A sound is comprised of two elements: • Envelope (outline of sound signal)

– Speech understanding in quiet • Fine structure (details of sound’s timing structure)

– Sound localization – Music perception – Speech understanding for tonal languages – Supports speech understanding in noise for Western languages

Envelope and Fine Structure A sound signal can also be divided into envelope and fine structure information  Envelope is the “loudness contour” and is essential for speech understanding (delivered by CIS)  Fine structure contains finest details of sound (rate) and enhances pitch and sound quality