Cochlear implant

     Hello. I want to know a lot about CIS algorithm that is used in cochlear implantation. If you could provide me any link related to that I would be happy. I would also like to know the code/ algorithm for Cochlear implantation tool in the Matlab.

[Daniel Shub]

Venky, it is much easier to contrbute if you make comments to the corresponding answer, edit your question where approriate and aske new questions when needed. You should also consider voting for answers that you think are helpful. This lets people know if they are on the right path.

Answer 1

[Wayne King]

Hi, You can read a nice introduction to the continuous interleaved sampling approach here

http://www.utdallas.edu/~loizou/cimplants/tutorial/

The DSP System Toolbox has a Simulink-based demo of a CIS algorithm in dspcochlear.mdl

[Wayne King]

I should say that the demo is missing the nonlinear compression that is typical in cochlear implants. The nonlinear compression is necessary because the tolerable dynamic range of electrical stimulation for patients (the difference between threshold and uncomfortable) is usually pretty small.

[i Venky]

Thank you Wayne King. If you find another link about this please tell me. Will read this one

Answer 2

[Daniel Shub]

Not sure where this should go. There are many different strategies for cochlear implant processing. I believe the two that are currently in use on new devices are CIS and SPEAK, with SPEAK begin used by Nucleus.

The F0/F2 is not CIS. At any one moment there are only two frequencies/features be provided to the user. These however rapidly change. If I recall correctly, F0 was encoded by the pulse rate and F2 by the electrode #. I think there may have been a loudness component coded by the amplitude of the pulses. One of the advantages of this approach is that only 1 electrode is active at any time.

For CIS all the electrodes, or a pre determined subset of them, are active. Each presents a pulse in sequence. The rate of the pulses and the pulse shape are fixed. The amplitude of the pulse train is modulated to provided information. Basically, the sound is split into a number of bands ~8-16 and the envelope is extracted. The bandpass filters and envelope extraction properties are usually proprietary. Researchers tend to use either halfwave rectification and low pass filtering or the Hilbert transform to get the envelope. Normal hearing models of CIS processing have been developed. There is a paper by Shannon (Science 1995) and another by Smith (Nature 2002) which present some interesting idea.

[i Venky]

I think I am getting it. There's only one band pass filter. It's range is 1000-6000 Hz and it's given 5 electrodes say each electrode has a range of 1000Hz and hence totally we have 5000Hz as the range. Now when you send the output of the band pass filter to the zero crossing detector we get the formant frequency. If the formant frequency is (say) 3300Hz then the 3rd electrode will be excited. The amplitude of the formant is the output of the envelope detector. The pulse rate is given by the fundamental frequency.

Is this right?

[Daniel Shub]

You are confusing F0/F2 with CIS. In CIS there are generally the same number of filters as there are electrodes. There are no zero-crossing detectors in CIS. The pulse rate in CIS is fixed. The output of each bandpass filter is feed to an evelope detector. The output of the envelope detector is used to modulate the pulse train.

Answer 3

[i Venky]

Hello Wayne. Where's that dspcochlear.mdl? Is that a file like .m file?

Thanks in advance.

[Wayne King]

Hi, .mdl is a Simulink model file. It is part of the DSP System Toolbox. If you search "Cochlear Implant Speech Processor" you will find it.

[i Venky]

Thanks

Answer 4

[i Venky]

Hello Wayne King. I read that paper. I couldn't understand F0/F2 technique (Nucleus Multi Electrode Implant). Could you please help me? If you have any link regarding this Nucleus Multi Electrode Implant please provide me the link.

Thanks in advance.

[i Venky]

Help me with this thing Wayne king.

Thanks.

[Wayne King]

Hi, all cochlear implant techniques for speech proceed from a desire to answer the question: what is the minimally sufficient information for speech understanding? The F0/F2 technique extracts the fundamental frequency envelope (amplitude modulation) and the second formant envelope (amplitude modulation) as important for speech understanding.

Answer 5

[i Venky]

I have only one doubt in that. F0/F2 involves two frequencies right? One is the fundamental and other one is the second formant. Then it should be mean that there should be only 2 electrodes right? But we still have many electrodes. That's the reason why I got confused. Thanks in advance.

[Wayne King]

Hi, that is correct, but that strategy alone is not used in any modern implant. If you look at the end of the article you will read about the MPEAK and SPEAK strategies. It's much more common now to use a filterbank with something like 8 to 16 bandpass filters. Of course, how many electrodes you can stimulate depends on the patient.

Answer 6

[i Venky]

 Okay I will see about MPEAK and SPEAK. How does zero crossing detector find out the formant?

Thanks in advance.

Answer 7

[Wayne King]

You want to bandpass filter of course, but then the zero crossing detector finds the dominant oscillation by noting that there are two zero crossings for every period of a sinusoid. So if you know the number of zero crossings in a given time interval you can estimate the frequency.

Answer 8

[i Venky]

 The output of the bandpass filter  contains many frequencies inbetween 300-1000Hz (say). Now if I pass this output of this filter to a zero crossing detector I will get the formant frequency within that range of frequencies (i.e. between 300-1000Hz). Am I right?

[i Venky]

It's still vague. I will ask doubts after I finish reading MPEAK and SPEAK.

Thank you. Will ask doubts soon.

[Wayne King]

That's right. You assume there is only one dominant mode of oscillation in that band.

Answer 9

[i Venky]

Okay. What's the algorithm that you use in Matlab for Cochlear?

I need the code.

[Wayne King]

The dspcochlear demo is a Simulink model, but if you have that software and enter

>>open_system('dspcochlear');

You can interact with the Simulink model and actually see how the CIS algorithm works with 8-band filter bank.

[i Venky]

Hello how to change the input? There are only two inputs- "chirp" and "speech signal". I need to give my own input. How should I do that?

Thanks in advance.

[Wayne King]

You can delete that block and use another source block in its place.

[i Venky]

Hello Wayne. I deleted that block and I selected Analog Input from the Data Acquisition Toolbox. But I am getting some kind of error that I couldn't resolve out. I think that the output data is not perfectly matched with that of the next stage. I would be really happy if you check that out and tell me the procedure fully to give some input to the system.