the computational neurophysics of perception

Started by maes...@ultrapiano.com October 29, 2005
Newsgroups:
rec.music.makers.piano,rec.music.makers.synth,rec.music.compose,alt.music,alt.sci.physics.acoustics
Subject: Re: how your ears work
Date: 25 Oct 2005 18:38:23 -0700

p...@aol.com wrote:
so what 'is' a sound?

maes...@ultrapiano.com wrote:
> Inside the cochlea of a human ear are hundreds of thin cilia. Rather > like an insect's eye, which is made up of hundreds of very primitive > eyes, each cilium responds individually to the pressure variations in > the cochlear fluid caused by sounds, contributing its own primitive > interpretation of the sound to the overall picture. Each cilium > responds to pressure by firing a synapse and sending a bio-electrical > signal along the aural nerve into the inner brain, the journey taking > about a 25th of a second. While a 25th of a second's worth of pressure > variation information is on the aural nerve's extensions into the brain, > the brain attempts to match its auditory memory with features in the > pressure variation information. Neural pathways into the brain's > memory of pressure variation patterns are strengthened during this > process, resulting in the brain deciding on what the sound 'is'.
>From the brain's point of view a sound 'is' approaching from the left,
for example, or a sound 'is' running water, or 'is' frightening, or pleasant. The brain can detect such qualities without thought, because of automatic biologically evolved processes occuring within it, such as comparison of remembered pressure variation patterns with those in the sound. Sound can be described mathematically as consisting of sinewaves at different frequencies and amplitudes, or as the numbers in a .wav or .mp3 file, but the information on the aural nerve is none of these. Each cilia fires when sufficient pressure above the ambient has been applied, and then takes about a 10th of a second to recover so that it is ready to fire again (by which time the internal pressure in the cochlea will usually have decreased to below the ambient for a short while). Some cilia are less responsive than others and take more time to recover, but each cilium individually behaves like a very primitive ear, which becomes dormant for a short while whenever it has been squashed by the pressure in the ear sufficiently for its synapse to fire. The hundreds of cilia individually send their primitive information along the aural nerve to the inner brain, and the brain can interpret the total of the sound information corresponding to a 25th of a second while it is on the aural nerve. A loud sudden sound would probably cause a lot of cilia to fire simultaneously, along with low-level echoes, and the brain can detect that pattern of information as it travels along the nerve fibres. High frequency sound might result in the frequent occurrence of large numbers of cilia firing in a short period of time, corresponding to a short length of the aural nerve on which the pattern is occurring. The often-repeated belief (sometimes called the travelling wave theory) that the ear itself hears sound mechanically (rather than the brain doing the hearing and deciding on what a sound 'is'), with each cilium tuned by nature to respond to a specific sinewave frequency, and vibrating sympathetically to 'sound waves' of that frequency because of resonance effects in the cochlear fluid, and the basilar membrane detecting the strength of the vibration (rather like the base of a cat's whisker) to give the amplitude of a frequency, is basically quite ridiculous and should be abandoned. Subject: Re: how your ears work Date: 27 Oct 2005 15:58:00 -0700 Matthew Fields wrote:
> In article <djradt$gu...@nyytiset.pp.htv.fi>, > J Ketutsalo <ei.kii...@spammia.fi> wrote: > >StpN...@aol.com wrote: > >> What the original "how your ears work" ( > >> http://ultrapiano.com/manufacturers/EarPage.jpg ) is getting at is that > >> the ear and brain do not perform spectral analysis of sound at all! > > > >Ok. I had to take a look at the text. It does not say that the ear and > >the brain do not perform spectral analysis. It says that the information > >in the auditory nerve is not a mere spectral analysis of sound, which is > >very much true. However, the neural impulses travelling through the > >auditory nerve *include* information about the spectral content of > >sound, and that information is used by the brain. No doubt about that. > > Or at least some sort of information from which spectral information > can be estimated.
Yes, of course the information that spectral analysis gives, is contained in the electrical impulses that the ear sends along the aural nerve into the brain. The information contained by the numbers in a .wav or .mp3 file is also contained in the same electrical impulses, but it is pointless to mention that as anything to do with what the brain does with the pressure variation information that it receives - spectral content is similarly irrelevant. It is also true that experiments have been performed with sinewave inputs to the ear, which cause indentifiable hair-cells to fire. However, these experiments do not claim to prove that specific hair-cells encode specific frequencies, but unfortunately they do not disprove it either. The hair-cells are all fairly similar, although the ones at the centre of the spiral cochlea are somewhat shorter and stubbier than the ones at the other end of the spiral. The only difference between them biologically is that the shorter haircells take more pressure before they fire their synapse, and also take longer to recover before they can fire again, than the longer ones. Because of that, each individual hair-cell must be responsive to all audible frequencies at any amplitude, but some are more responsive than others. The ear/brain is actually tuned 'by nature' to hear and understand human speech, and is particularly good at identifying the tiny nuances which give the emotional content and indicate the speaker's state of mind. The ups and downs in pitch and the changes in loudness during normal speech are all part of what the brain recognises when it hears a voice, but does anyone suggest that the ear is encoding these important parts of spoken sound at the hair-cell level? Of course not, the hair-like structures have not evolved to detect 'ups' or 'downs' in the voice of a potential mate - so why should anyone take seriously the idea that the ear performs a spectral analysis of human speech - what evolutionary advantage would it give? If someone had only ever heard their parents talking, and had never heard sinewaves before, if you played a sinewave (440Hz from a tuning-fork A, for example) they would at first say that the tuning fork sounded like their mother or father, until they had learned that the new sound was a different thing. The ability to identify sinewaves is a learnt ability, and is not evolved by nature. I think that most theorists are confusing themselves with the idea of building an 'artifical ear' that can hear by itself, which could be perhaps plugged into the brain to cure deafness. Human ears don't hear - they channel air pressure variations into the cochlea. The hair-cells transduce these pressure variations into electrical impulses which travel along the aural nerve. The aural nerve acts as a delay line, containing about a 25th of a second's worth of information at any one time. The brain matches features and patterns in the information on the aural nerve against remembered information patterns (from previously heard sounds) to produce the sensation of hearing. Subject: Re: how your ears work Date: 28 Oct 2005 16:02:01 -0700 J Ketutsalo wrote:
> StpN...@aol.com wrote: > > > > I think that most theorists are confusing themselves with the idea of > > building an 'artifical ear' that can hear by itself, which could be > > perhaps plugged into the brain to cure deafness. > > I don't know any theorist who would claim things are that simple.
By 'theorists' I include 'conspiracy theorists', who claim that brain research has reached the stage where governments control us all with mind implants etc. - I think they are confused.
> > Human ears don't hear > > - they channel air pressure variations into the cochlea. The > > hair-cells transduce these pressure variations into electrical impulses > > which travel along the aural nerve. The aural nerve acts as a delay > > line, containing about a 25th of a second's worth of information at any > > one time. > > The auditory nerve does not preserve any information. True, it takes a > finite time for the sound to travel through the nerve, but the brain > cannot sample what is in the middle. 25th of a second sounds like a time > when the information is well beyond the auditory nerve but has not > reached the cortex, yet. At these intermediate stages some past > information is available. > > > The brain matches features and patterns in the information > > on the aural nerve against remembered information patterns (from > > previously heard sounds) to produce the sensation of hearing. > > On this I can agree.
The explanation I gave is of necessity simplified, and I hope the simplified terminology I used is not too misleading. I was calling the hair-cells 'cilia' until someone wrote and told me that that is how researchers refer to the 'spirocilia', which are little tufts of tiny hairs at the end of each hair-cell - in the literature, these are apparently sometimes confused with the organ of Corti, which in turn is sometimes confused with the basilar membrane. Sometimes, pictures of the 'rods and cones' in the eye have been displayed on internet websites and labelled as pictures of the ear's 'cilia'. You may well be right, that the auditory nerve only holds information temporarily, that the brain "cannot sample what is in the middle" [of the auditory nerve], and that after a 25th of a second any information from the ear will have travelled beyond the auditory nerve towards the auditory cortex. The explanation I gave is still essentially correct, however, in that the brain keeps about a 25th of a second's worth of input sound pressure information at any one time, on a delay-line principle. The information does not necessarily all have to be on the auditory nerve, but can be spread out, beyond what you might define as the end of the auditory nerve, into the neural pathways which extend further into the brain. We do of course have two ears, and the brain is split into left and right halves, connected by a thick bundle of fibres for communication between the two halves. When hearing, the most basic function performed by the brain, is comparison of the information from the left and right ear to detect where a sound is coming from, and at the same time the brain tries to match this information with previously heard sounds to identify what it is hearing. The left brain handles the information from the right ear, and vice versa. It is not unreasonable to suppose that the brain preserves both versions of the information it hears, to make subsequent matching more efficient. Subject: Re: how your ears work Date: Thu, 27 Oct 2005 01:37:35 GMT In article <1130373939.164331.24...@g14g2000cwa.googlegroups.com>, <StpN...@aol.com> wrote:
>S.O.D.D.I. wrote: >> J Ketutsalo wrote: >> > What is this conversation doing in the piano and synth groups? >> >> I dunno about the piano group, but sound perception (and anomalies) are >> very interesting to this synthesist. > >It was something to do with the statement about 25ths of seconds in the >thread "special effects from pianos": > >http://groups.google.co.uk/group/rec.music.compose/msg/4f965c3c2b89ae5?hl=en > >"On a normal piano, the maximum repeat rate of an individual note is >about ten per second, but If a piano note could be repeated at the rate >of about 25 notes per second, the result would be a single continuous >tone." > >Did you know that if you split up a sound into 25ths of a second and >then play each 25th of a second backwards consecutively, it sounds the >same as the original? (other from any clickiness caused by the joins). > > >It occurred to me that that might from the basis for a DSP data >compression algorithm - if you play each 25th of a second forwards and >backwards at the same time, the resultant signal is symmetrical, so you >only need to store half the signal! Isn't it amazing what clever ears >we have - no matter what is fed into them, forwards, backwards >inside-out or upside-down, our brains can always make some sense of it! > - or is that because we are basically stupid?
This is already built into MP3. What's more interesting is that back in the dark ages of computers, a fellow called Iannis Xenakis proposed making entirely new sounds by splicing together segments of roughly this length, and then another much younger fellow (about my age) by the name of Xavier Serra realized that by taking segments this length and overlapping them... or taking similar but overlapping segments of an original sound and reducing the amount of overlap, he could speed up or slow down a sound without changing its pitch. He quickly discovered the use of something called Hamming's Window which eliminated most of the white noise of the splicing clicks. These algorithms were widely published and are built into many DSP systems today, including many you may be familiar with. Subject: Re: how your ears work Date: 26 Oct 2005 19:30:48 -0700 MP3 and Hamming windows compression are mostly related to spectral analysis of sound into its component frequencies and then doing clever mathematical transformations of the results. My simple approach of playing each 25th of a second backwards and forwards at the same time requires hardly any processing, other than the de-clicking. The clicks are quite predictable and easy to understand, so they can quite easily be eliminated - by readjusting the zero at each splice, for example, or selecting a convenient splice point. The clicks aren't white noise, they are caused by the loudspeaker cone attempting to move or change direction too quickly for its design specifications - some digitally-generated clicks are actually miniature 'sonic booms' caused by the loudspeaker diaphragm moving faster than the speed of sound! (If the the maximum distance that it can move is 1/4", if consecutive samples are the max and min values, at 40KHz sampling rate the speed is 10,000"/sec). From: StpN...@aol.com Newsgroups: rec.music.makers.piano,rec.music.makers.synth,rec.music.compose,alt.music,alt.sci.physics.acoustics Subject: Re: how your ears work Date: 27 Oct 2005 11:25:36 -0700 Here's a short program you can use to test .wav files (It's untidy and badly written but it works! - the first chunk handles the .wav formatting information and then it just reverses every 25th of a second of the input file and puts it in the output file). Matthew Fields wrote:
> I'd be interested in how well this reversing 1/25s of a second works > for high unreverberated xylophone notes. I choose those because > almost-periodicity sets in almost immediately after the bars are struck, > and the amplitude tends to decay to inaudble within a single cycle--which > indicates just how sensitive our frequency-extrapolating mechanism is, but > it also means you may have much less than 1sec/25 to work with. >
#include <io.h> #include <stdio.h> /* main algorithm opens INfile reads 25ths of seconds forwards, writes backwards to OUTfile */ #define BUFFSIZE 1764 /* 25th of a second at the sampling rate of INFile */ FILE *OUTfile, *INfile; char ChunkID[4]; long LongInt; short int ShortInt; short int n, buffer[BUFFSIZE], bcount; int main(int argc, char **argv) { if((INfile = fopen("C:\\sound\\wavs\\INmail.wav","rb"))==NULL) { printf("failed to open file C:\\sound\\wavs\\INmail\n"); } else printf("Input file C:\\sound\\wavs\\INmail.wav opened OK\n"); if((OUTfile = fopen("C:\\sound\\wavs\\OUTmail.wav","wb"))==NULL) { printf("failed to open file C:\\sound\\wavs\\OUTmail.wav\n"); } else printf("Output file C:\\sound\\wavs\\OUTmail.wav opened OK\n"); n = fread(ChunkID, 4, 1, INfile); /* "RIFF" */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* Total INfile size (-4) */ printf("Chunksize is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "WAVE" */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "fmt " */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* 16 */ printf("SubChunk1Size is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* 1 */ printf("AudioFormat is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* 1 Mono, 2 Stereo */ printf("NumChannels is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* sample rate */ printf("SampleRate is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* byte rate */ printf("ByteRate is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* */ printf("BlockAlign is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* */ printf("BitsPerSample is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "data" */ printf(" %c %c %c %c\n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* */ printf("SubChunk2Size is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); bcount=0; while (LongInt > 0) { LongInt--; LongInt--; n = fread(&ShortInt, 2, 1, INfile); buffer[bcount] = ShortInt; bcount++; if (bcount == BUFFSIZE) { while (bcount > 0) { bcount--; ShortInt = buffer[BUFFSIZE-bcount]; fwrite(&ShortInt, 2, 1, OUTfile); } } } while (bcount) { fwrite(&ShortInt, 2, 1, OUTfile); bcount--; } fclose(INfile); fclose(OUTfile); return(0); }
in article 1130625124.508354.259710@g43g2000cwa.googlegroups.com,
maestro@ultrapiano.com at StpNrrs@aol.com wrote on 10/29/2005 18:32:

> Newsgroups: > rec.music.makers.piano,rec.music.makers.synth,rec.music.compose,alt.music,alt. > sci.physics.acoustics > Subject: Re: how your ears work > Date: 25 Oct 2005 18:38:23 -0700 > > p...@aol.com wrote: > so what 'is' a sound? > > maes...@ultrapiano.com wrote: >> Inside the cochlea of a human ear are hundreds of thin cilia. Rather >> like an insect's eye, which is made up of hundreds of very primitive >> eyes, each cilium responds individually to the pressure variations in >> the cochlear fluid caused by sounds, contributing its own primitive >> interpretation of the sound to the overall picture. Each cilium >> responds to pressure by firing a synapse and sending a bio-electrical >> signal along the aural nerve into the inner brain, the journey taking >> about a 25th of a second. While a 25th of a second's worth of pressure >> variation information is on the aural nerve's extensions into the brain, >> the brain attempts to match its auditory memory with features in the >> pressure variation information. Neural pathways into the brain's >> memory of pressure variation patterns are strengthened during this >> process, resulting in the brain deciding on what the sound 'is'. > > > From the brain's point of view a sound 'is' approaching from the left, > for example, or a sound 'is' running water, or 'is' frightening, or > pleasant. The brain can detect such qualities without thought, because > of automatic biologically evolved processes occuring within it, such as > comparison of remembered pressure variation patterns with those in the > sound. > > Sound can be described mathematically as consisting of sinewaves at > different frequencies and amplitudes, or as the numbers in a .wav or > .mp3 file, but the information on the aural nerve is none of these. > Each cilia fires when sufficient pressure above the ambient has been > applied, and then takes about a 10th of a second to recover so that it > is ready to fire again (by which time the internal pressure in the > cochlea will usually have decreased to below the ambient for a short > while). Some cilia are less responsive than others and take more time > to recover, but each cilium individually behaves like a very primitive > ear, which becomes dormant for a short while whenever it has been > squashed by the pressure in the ear sufficiently for its synapse to > fire. > > The hundreds of cilia individually send their primitive information > along the aural nerve to the inner brain, and the brain can interpret > the total of the sound information corresponding to a 25th of a second > while it is on the aural nerve. A loud sudden sound would probably > cause a lot of cilia to fire simultaneously, along with low-level > echoes, and the brain can detect that pattern of information as it > travels along the nerve fibres. High frequency sound might result in > the frequent occurrence of large numbers of cilia firing in a short > period of time, corresponding to a short length of the aural nerve on > which the pattern is occurring. > > The often-repeated belief (sometimes called the travelling wave theory) > that the ear itself hears sound mechanically (rather than the brain > doing the hearing and deciding on what a sound 'is'), with each cilium > tuned by nature to respond to a specific sinewave frequency, and > vibrating sympathetically to 'sound waves' of that frequency because of > resonance effects in the cochlear fluid, and the basilar membrane > detecting the strength of the vibration (rather like the base of a > cat's whisker) to give the amplitude of a frequency, is basically quite > ridiculous and should be abandoned.
i hadn't read through all this (and not guaranteeing that i will) but a *very* interesting paper presented by a guy how designs signal processing chips, some doing specific noise-shaping processing, was presented at IEEE Mohonk 1997: http://www.icsi.berkeley.edu/~dpwe/waspaa97/program.html it was Bob Adams (AD1890, ADI "sigma DSP", etc) "Spectral noise-shaping in integrate-and-fire neural networks". he also presented it the earlier EAS convention in LA. i saw both and thought it was a compelling theory. anybody know an on-line source for the paper? -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
in article BF8973FB.BAB2%rbj@audioimagination.com, robert bristow-johnson at
rbj@audioimagination.com wrote on 10/29/2005 18:46:

> > i hadn't read through all this (and not guaranteeing that i will) but a > *very* interesting paper presented by a guy how designs signal processing > chips, some doing specific noise-shaping processing, was presented at IEEE > Mohonk 1997: http://www.icsi.berkeley.edu/~dpwe/waspaa97/program.html > > it was Bob Adams (AD1890, ADI "sigma DSP", etc) "Spectral noise-shaping in > integrate-and-fire neural networks". he also presented it the earlier EAS > convention in LA. i saw both and thought it was a compelling theory. > anybody know an on-line source for the paper? >
duh, i googled the entire title and found a legit way to get the paper here: http://ieeexplore.ieee.org/search/wrapper.jsp?arnumber=618072 but i was looking for a freebie. the abstract is: Spectral noise-shaping in integrate-and-fire neural networks Adams, R.W.; Neural Networks,1997., International Conference on Volume 2,&#2013266080; 9-12 June 1997 Page(s):953 - 958 vol.2 Abstract: A theory of coordinated neuronal firing events is proposed that allows the low-noise transmission of analog signals through a network of coupled neurons. The inherently high noise levels of a biological neuron can be reduced over a specified frequency range using a network of interconnected neurons. These interconnections cause certain statistical temporal patterns to occur in the summed output of a modest number of neurons, and these temporal patterns can dramatically improve the signal-to-noise ratio over a given frequency range -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
A sort of statistical neurodynamics of noise transmission, I suppose.
What practical use is your friend's (Bob Adams) theory?  Does it
explain the brain's equivalent to Brownian motion - those odd sudden
jolts and tics that one sometimes experiences when half-asleep?   I
found the 1997 paper you are discussing rather enervating - why not
read what I wrote and reply to that instead?

robert bristow-johnson wrote:
> in article BF8973FB.BAB2%rbj@audioimagination.com, robert bristow-johnson at > rbj@audioimagination.com wrote on 10/29/2005 18:46: > > > > > i hadn't read through all this (and not guaranteeing that i will) but a > > *very* interesting paper presented by a guy how designs signal processing > > chips, some doing specific noise-shaping processing, was presented at IEEE > > Mohonk 1997: http://www.icsi.berkeley.edu/~dpwe/waspaa97/program.html > > > > it was Bob Adams (AD1890, ADI "sigma DSP", etc) "Spectral noise-shaping in > > integrate-and-fire neural networks". he also presented it the earlier EAS > > convention in LA. i saw both and thought it was a compelling theory. > > anybody know an on-line source for the paper? > > > > duh, i googled the entire title and found a legit way to get the paper here: > http://ieeexplore.ieee.org/search/wrapper.jsp?arnumber=618072 > but i was looking for a freebie. the abstract is: > > Spectral noise-shaping in integrate-and-fire neural networks > Adams, R.W.; > Neural Networks,1997., International Conference on > Volume 2, 9-12 June 1997 Page(s):953 - 958 vol.2 > > Abstract: > > A theory of coordinated neuronal firing events is proposed that allows the > low-noise transmission of analog signals through a network of coupled > neurons. The inherently high noise levels of a biological neuron can be > reduced over a specified frequency range using a network of interconnected > neurons. These interconnections cause certain statistical temporal patterns > to occur in the summed output of a modest number of neurons, and these > temporal patterns can dramatically improve the signal-to-noise ratio over a > given frequency range > > > -- > > r b-j rbj@audioimagination.com > > "Imagination is more important than knowledge."
Newsgroups: rec.music.makers.piano,rec.music.makers.synth,rec.music.compose,alt.music,alt.sci.physics.acoustics Subject: Re: how your ears work Date: 25 Oct 2005 18:38:23 -0700 p...@aol.com wrote: so what 'is' a sound? maes...@ultrapiano.com wrote:
> Inside the cochlea of a human ear are hundreds of thin cilia. Rather > like an insect's eye, which is made up of hundreds of very primitive > eyes, each cilium responds individually to the pressure variations in > the cochlear fluid caused by sounds, contributing its own primitive > interpretation of the sound to the overall picture. Each cilium > responds to pressure by firing a synapse and sending a bio-electrical > signal along the aural nerve into the inner brain, the journey taking > about a 25th of a second. While a 25th of a second's worth of pressure > variation information is on the aural nerve's extensions into the brain, > the brain attempts to match its auditory memory with features in the > pressure variation information. Neural pathways into the brain's > memory of pressure variation patterns are strengthened during this > process, resulting in the brain deciding on what the sound 'is'.
>From the brain's point of view a sound 'is' approaching from the left,
for example, or a sound 'is' running water, or 'is' frightening, or pleasant. The brain can detect such qualities without thought, because of automatic biologically evolved processes occuring within it, such as comparison of remembered pressure variation patterns with those in the sound. Sound can be described mathematically as consisting of sinewaves at different frequencies and amplitudes, or as the numbers in a .wav or .mp3 file, but the information on the aural nerve is none of these. Each cilia fires when sufficient pressure above the ambient has been applied, and then takes about a 10th of a second to recover so that it is ready to fire again (by which time the internal pressure in the cochlea will usually have decreased to below the ambient for a short while). Some cilia are less responsive than others and take more time to recover, but each cilium individually behaves like a very primitive ear, which becomes dormant for a short while whenever it has been squashed by the pressure in the ear sufficiently for its synapse to fire. The hundreds of cilia individually send their primitive information along the aural nerve to the inner brain, and the brain can interpret the total of the sound information corresponding to a 25th of a second while it is on the aural nerve. A loud sudden sound would probably cause a lot of cilia to fire simultaneously, along with low-level echoes, and the brain can detect that pattern of information as it travels along the nerve fibres. High frequency sound might result in the frequent occurrence of large numbers of cilia firing in a short period of time, corresponding to a short length of the aural nerve on which the pattern is occurring. The often-repeated belief (sometimes called the travelling wave theory) that the ear itself hears sound mechanically (rather than the brain doing the hearing and deciding on what a sound 'is'), with each cilium tuned by nature to respond to a specific sinewave frequency, and vibrating sympathetically to 'sound waves' of that frequency because of resonance effects in the cochlear fluid, and the basilar membrane detecting the strength of the vibration (rather like the base of a cat's whisker) to give the amplitude of a frequency, is basically quite ridiculous and should be abandoned. Subject: Re: how your ears work Date: 27 Oct 2005 15:58:00 -0700 Matthew Fields wrote:
> In article <djradt$gu...@nyytiset.pp.htv.fi>, > J Ketutsalo <ei.kii...@spammia.fi> wrote: > >StpN...@aol.com wrote: > >> What the original "how your ears work" ( > >> http://ultrapiano.com/manufacturers/EarPage.jpg ) is getting at is that > >> the ear and brain do not perform spectral analysis of sound at all! > > > >Ok. I had to take a look at the text. It does not say that the ear and > >the brain do not perform spectral analysis. It says that the information > >in the auditory nerve is not a mere spectral analysis of sound, which is > >very much true. However, the neural impulses travelling through the > >auditory nerve *include* information about the spectral content of > >sound, and that information is used by the brain. No doubt about that. > > Or at least some sort of information from which spectral information > can be estimated.
Yes, of course the information that spectral analysis gives, is contained in the electrical impulses that the ear sends along the aural nerve into the brain. The information contained by the numbers in a .wav or .mp3 file is also contained in the same electrical impulses, but it is pointless to mention that as anything to do with what the brain does with the pressure variation information that it receives - spectral content is similarly irrelevant. It is also true that experiments have been performed with sinewave inputs to the ear, which cause indentifiable hair-cells to fire. However, these experiments do not claim to prove that specific hair-cells encode specific frequencies, but unfortunately they do not disprove it either. The hair-cells are all fairly similar, although the ones at the centre of the spiral cochlea are somewhat shorter and stubbier than the ones at the other end of the spiral. The only difference between them biologically is that the shorter haircells take more pressure before they fire their synapse, and also take longer to recover before they can fire again, than the longer ones. Because of that, each individual hair-cell must be responsive to all audible frequencies at any amplitude, but some are more responsive than others. The ear/brain is actually tuned 'by nature' to hear and understand human speech, and is particularly good at identifying the tiny nuances which give the emotional content and indicate the speaker's state of mind. The ups and downs in pitch and the changes in loudness during normal speech are all part of what the brain recognises when it hears a voice, but does anyone suggest that the ear is encoding these important parts of spoken sound at the hair-cell level? Of course not, the hair-like structures have not evolved to detect 'ups' or 'downs' in the voice of a potential mate - so why should anyone take seriously the idea that the ear performs a spectral analysis of human speech - what evolutionary advantage would it give? If someone had only ever heard their parents talking, and had never heard sinewaves before, if you played a sinewave (440Hz from a tuning-fork A, for example) they would at first say that the tuning fork sounded like their mother or father, until they had learned that the new sound was a different thing. The ability to identify sinewaves is a learnt ability, and is not evolved by nature. I think that most theorists are confusing themselves with the idea of building an 'artifical ear' that can hear by itself, which could be perhaps plugged into the brain to cure deafness. Human ears don't hear - they channel air pressure variations into the cochlea. The hair-cells transduce these pressure variations into electrical impulses which travel along the aural nerve. The aural nerve acts as a delay line, containing about a 25th of a second's worth of information at any one time. The brain matches features and patterns in the information on the aural nerve against remembered information patterns (from previously heard sounds) to produce the sensation of hearing. Subject: Re: how your ears work Date: 28 Oct 2005 16:02:01 -0700 J Ketutsalo wrote:
> StpN...@aol.com wrote: > > > > I think that most theorists are confusing themselves with the idea of > > building an 'artifical ear' that can hear by itself, which could be > > perhaps plugged into the brain to cure deafness. > > I don't know any theorist who would claim things are that simple.
By 'theorists' I include 'conspiracy theorists', who claim that brain research has reached the stage where governments control us all with mind implants etc. - I think they are confused.
> > Human ears don't hear > > - they channel air pressure variations into the cochlea. The > > hair-cells transduce these pressure variations into electrical impulses > > which travel along the aural nerve. The aural nerve acts as a delay > > line, containing about a 25th of a second's worth of information at any > > one time. > > The auditory nerve does not preserve any information. True, it takes a > finite time for the sound to travel through the nerve, but the brain > cannot sample what is in the middle. 25th of a second sounds like a time > when the information is well beyond the auditory nerve but has not > reached the cortex, yet. At these intermediate stages some past > information is available. > > > The brain matches features and patterns in the information > > on the aural nerve against remembered information patterns (from > > previously heard sounds) to produce the sensation of hearing. > > On this I can agree.
The explanation I gave is of necessity simplified, and I hope the simplified terminology I used is not too misleading. I was calling the hair-cells 'cilia' until someone wrote and told me that that is how researchers refer to the 'spirocilia', which are little tufts of tiny hairs at the end of each hair-cell - in the literature, these are apparently sometimes confused with the organ of Corti, which in turn is sometimes confused with the basilar membrane. Sometimes, pictures of the 'rods and cones' in the eye have been displayed on internet websites and labelled as pictures of the ear's 'cilia'. You may well be right, that the auditory nerve only holds information temporarily, that the brain "cannot sample what is in the middle" [of the auditory nerve], and that after a 25th of a second any information from the ear will have travelled beyond the auditory nerve towards the auditory cortex. The explanation I gave is still essentially correct, however, in that the brain keeps about a 25th of a second's worth of input sound pressure information at any one time, on a delay-line principle. The information does not necessarily all have to be on the auditory nerve, but can be spread out, beyond what you might define as the end of the auditory nerve, into the neural pathways which extend further into the brain. We do of course have two ears, and the brain is split into left and right halves, connected by a thick bundle of fibres for communication between the two halves. When hearing, the most basic function performed by the brain, is comparison of the information from the left and right ear to detect where a sound is coming from, and at the same time the brain tries to match this information with previously heard sounds to identify what it is hearing. The left brain handles the information from the right ear, and vice versa. It is not unreasonable to suppose that the brain preserves both versions of the information it hears, to make subsequent matching more efficient. Subject: Re: how your ears work Date: Thu, 27 Oct 2005 01:37:35 GMT In article <1130373939.164331.24...@g14g2000cwa.googlegroups.com>, <StpN...@aol.com> wrote:
>S.O.D.D.I. wrote: >> J Ketutsalo wrote: >> > What is this conversation doing in the piano and synth groups? >> >> I dunno about the piano group, but sound perception (and anomalies) are >> very interesting to this synthesist. > >It was something to do with the statement about 25ths of seconds in the >thread "special effects from pianos": > >http://groups.google.co.uk/group/rec.music.compose/msg/4f965c3c2b89ae5?hl=en > >"On a normal piano, the maximum repeat rate of an individual note is >about ten per second, but If a piano note could be repeated at the rate >of about 25 notes per second, the result would be a single continuous >tone." > >Did you know that if you split up a sound into 25ths of a second and >then play each 25th of a second backwards consecutively, it sounds the >same as the original? (other from any clickiness caused by the joins). > > >It occurred to me that that might from the basis for a DSP data >compression algorithm - if you play each 25th of a second forwards and >backwards at the same time, the resultant signal is symmetrical, so you >only need to store half the signal! Isn't it amazing what clever ears >we have - no matter what is fed into them, forwards, backwards >inside-out or upside-down, our brains can always make some sense of it! > - or is that because we are basically stupid?
This is already built into MP3. What's more interesting is that back in the dark ages of computers, a fellow called Iannis Xenakis proposed making entirely new sounds by splicing together segments of roughly this length, and then another much younger fellow (about my age) by the name of Xavier Serra realized that by taking segments this length and overlapping them... or taking similar but overlapping segments of an original sound and reducing the amount of overlap, he could speed up or slow down a sound without changing its pitch. He quickly discovered the use of something called Hamming's Window which eliminated most of the white noise of the splicing clicks. These algorithms were widely published and are built into many DSP systems today, including many you may be familiar with. Subject: Re: how your ears work Date: 26 Oct 2005 19:30:48 -0700 MP3 and Hamming windows compression are mostly related to spectral analysis of sound into its component frequencies and then doing clever mathematical transformations of the results. My simple approach of playing each 25th of a second backwards and forwards at the same time requires hardly any processing, other than the de-clicking. The clicks are quite predictable and easy to understand, so they can quite easily be eliminated - by readjusting the zero at each splice, for example, or selecting a convenient splice point. The clicks aren't white noise, they are caused by the loudspeaker cone attempting to move or change direction too quickly for its design specifications - some digitally-generated clicks are actually miniature 'sonic booms' caused by the loudspeaker diaphragm moving faster than the speed of sound! (If the the maximum distance that it can move is 1/4", if consecutive samples are the max and min values, at 40KHz sampling rate the speed is 10,000"/sec). From: StpN...@aol.com Newsgroups: rec.music.makers.piano,rec.music.makers.synth,rec.music.compose,alt.music,alt.sci.physics.acoustics Subject: Re: how your ears work Date: 27 Oct 2005 11:25:36 -0700 Here's a short program you can use to test .wav files (It's untidy and badly written but it works! - the first chunk handles the .wav formatting information and then it just reverses every 25th of a second of the input file and puts it in the output file). Matthew Fields wrote:
> I'd be interested in how well this reversing 1/25s of a second works > for high unreverberated xylophone notes. I choose those because > almost-periodicity sets in almost immediately after the bars are struck, > and the amplitude tends to decay to inaudble within a single cycle--which > indicates just how sensitive our frequency-extrapolating mechanism is, but > it also means you may have much less than 1sec/25 to work with. >
#include <io.h> #include <stdio.h> /* main algorithm opens INfile reads 25ths of seconds forwards, writes backwards to OUTfile */ #define BUFFSIZE 1764 /* 25th of a second at the sampling rate of INFile */ FILE *OUTfile, *INfile; char ChunkID[4]; long LongInt; short int ShortInt; short int n, buffer[BUFFSIZE], bcount; int main(int argc, char **argv) { if((INfile = fopen("C:\\sound\\wavs\\INmail.wav","rb"))==NULL) { printf("failed to open file C:\\sound\\wavs\\INmail\n"); } else printf("Input file C:\\sound\\wavs\\INmail.wav opened OK\n"); if((OUTfile = fopen("C:\\sound\\wavs\\OUTmail.wav","wb"))==NULL) { printf("failed to open file C:\\sound\\wavs\\OUTmail.wav\n"); } else printf("Output file C:\\sound\\wavs\\OUTmail.wav opened OK\n"); n = fread(ChunkID, 4, 1, INfile); /* "RIFF" */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* Total INfile size (-4) */ printf("Chunksize is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "WAVE" */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "fmt " */ printf(" %c %c %c %c \n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* 16 */ printf("SubChunk1Size is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* 1 */ printf("AudioFormat is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* 1 Mono, 2 Stereo */ printf("NumChannels is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* sample rate */ printf("SampleRate is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* byte rate */ printf("ByteRate is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* */ printf("BlockAlign is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(&ShortInt, 2, 1, INfile); /* */ printf("BitsPerSample is %d\n", ShortInt); fwrite(&ShortInt, 2, 1, OUTfile); n = fread(ChunkID, 4, 1, INfile); /* "data" */ printf(" %c %c %c %c\n", ChunkID[0],ChunkID[1],ChunkID[2],ChunkID[3]); fwrite(ChunkID, 4, 1, OUTfile); n = fread(&LongInt, 4, 1, INfile); /* */ printf("SubChunk2Size is %d\n", LongInt); fwrite(&LongInt, 4, 1, OUTfile); bcount=0; while (LongInt > 0) { LongInt--; LongInt--; n = fread(&ShortInt, 2, 1, INfile); buffer[bcount] = ShortInt; bcount++; if (bcount == BUFFSIZE) { while (bcount > 0) { bcount--; ShortInt = buffer[BUFFSIZE-bcount]; fwrite(&ShortInt, 2, 1, OUTfile); } } } while (bcount) { fwrite(&ShortInt, 2, 1, OUTfile); bcount--; } fclose(INfile); fclose(OUTfile); return(0); }
in article 1130629424.117822.102680@f14g2000cwb.googlegroups.com,
maestro@ultrapiano.com at StpNrrs@aol.com wrote on 10/29/2005 19:43:

> A sort of statistical neurodynamics of noise transmission, I suppose. > What practical use is your friend's (Bob Adams) theory? Does it > explain the brain's equivalent to Brownian motion - those odd sudden > jolts and tics that one sometimes experiences when half-asleep?
he was able to explain, using the audio engineering concept of noise-shaping, how it is that we hear sound so well, to such a "fidelity" that using silicon-based systems (44100*16 bits per second in one ear), through this known neural bottleneck of our auditory nervous system.
> I found the 1997 paper you are discussing rather enervating - why not > read what I wrote and reply to that instead?
i'm getting there. there was a lot of stuff suddenly plopped onto comp.dsp. i haven't caught up with the entire thing posted nor with the existing thread from wherever this came from. still pretty busy. expect sporadic response. -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
robert bristow-johnson wrote:
> in article 1130629424.117822.102680@f14g2000cwb.googlegroups.com, > maestro@ultrapiano.com at StpNrrs@aol.com wrote on 10/29/2005 19:43: > > > A sort of statistical neurodynamics of noise transmission, I suppose. > > What practical use is your friend's (Bob Adams) theory? Does it > > explain the brain's equivalent to Brownian motion - those odd sudden > > jolts and tics that one sometimes experiences when half-asleep? > > he was able to explain, using the audio engineering concept of > noise-shaping, how it is that we hear sound so well, to such a "fidelity" > that using silicon-based systems (44100*16 bits per second in one ear), > through this known neural bottleneck of our auditory nervous system.
"this known neural bottleneck". Do you mean the disparity in the amount of information that the ear sends the brain, at most a few thousand haircells firing ten times per second, compared to the 44100*16 bits per second required for high-fidelity digital audio sound? The same 'known neural bottleneck' occurs in the eyes - about 1000000*3*25 bits per second when your're looking at your computer monitor.
"maestro@ultrapiano.com" <StpNrrs@aol.com> wrote in message 
news:1130625124.508354.259710@g43g2000cwa.googlegroups.com...

<snip>

> > The often-repeated belief (sometimes called the travelling wave theory) > that the ear itself hears sound mechanically (rather than the brain > doing the hearing and deciding on what a sound 'is'), with each cilium > tuned by nature to respond to a specific sinewave frequency, and > vibrating sympathetically to 'sound waves' of that frequency because of > resonance effects in the cochlear fluid, and the basilar membrane > detecting the strength of the vibration (rather like the base of a > cat's whisker) to give the amplitude of a frequency, is basically quite > ridiculous and should be abandoned.
LOL. <snip>
> but it is pointless to mention that as anything to do with what the > brain does with the pressure variation information that it receives - > spectral content is similarly irrelevant.
Hillarious. <snip>
> > The ear/brain is actually tuned 'by nature' to hear and understand > human speech,
Whoa. Too much detail. <snip>
> > If someone had only ever heard their parents talking, and had never > heard sinewaves before, if you played a sinewave (440Hz from a > tuning-fork A, for example) they would at first say that the tuning > fork sounded like their mother or father,
Funny but I don't remember ever misidentifying a tuning as a person.
> until they had learned that > the new sound was a different thing. The ability to identify sinewaves > is a learnt ability, and is not evolved by nature.
Ah yes. I fondly remember my third grade sinewave identifying teacher. <snip>
> selecting a convenient splice point. The clicks aren't white noise, > they are caused by the loudspeaker cone attempting to move or change > direction too quickly for its design specifications - some > digitally-generated clicks are actually miniature 'sonic booms' caused > by the loudspeaker diaphragm moving faster than the speed of sound!
Wow! That's a hell of a speaker! Looking at your "ultrapiano" page, it is a load of nonsense. No such thing has been built or ever will be built. You don't have the knowledge of physics, acoustics, or signals to know that it can't work as you describe it. -- rb
in article 1130632469.454161.195380@o13g2000cwo.googlegroups.com,
StpNrrs@aol.com at StpNrrs@aol.com wrote on 10/29/2005 20:34:

> > robert bristow-johnson wrote: >> in article 1130629424.117822.102680@f14g2000cwb.googlegroups.com, >> maestro@ultrapiano.com at StpNrrs@aol.com wrote on 10/29/2005 19:43: >> >>> A sort of statistical neurodynamics of noise transmission, I suppose. >>> What practical use is your friend's (Bob Adams) theory? Does it >>> explain the brain's equivalent to Brownian motion - those odd sudden >>> jolts and tics that one sometimes experiences when half-asleep? >> >> he was able to explain, using the audio engineering concept of >> noise-shaping, how it is that we hear sound so well, to such a "fidelity" >> that using silicon-based systems (44100*16 bits per second in one ear), >> through this known neural bottleneck of our auditory nervous system. > > "this known neural bottleneck". Do you mean the disparity in the > amount of information that the ear sends the brain, at most a few > thousand haircells firing ten times per second, compared to the > 44100*16 bits per second required for high-fidelity digital audio > sound?
yup. that's just about right. Bob Adams's paper deals with this in an elegant and quite believable manner. i'm not saying that is error free or has considered every issue, but it's pretty good.
> The same 'known neural bottleneck' occurs in the eyes - about > 1000000*3*25 bits per second when your're looking at your computer > monitor.
that may be. i was only thinking about the audiology part of "the computational neurophysics of perception". -- r b-j rbj@audioimagination.com "Imagination is more important than knowledge."
Ron Baker,    Pluralitas! wrote:
> "maestro@ultrapiano.com" <StpNrrs@aol.com> wrote in message > news:1130625124.508354.259710@g43g2000cwa.googlegroups.com... > > <snip> > > LOL. > > <snip>> > > Hillarious. > > <snip> > > Whoa. Too much detail. > > <snip> > > Funny but I don't remember ever misidentifying a tuning > as a person. > > Ah yes. I fondly remember my third grade sinewave > identifying teacher. > > <snip> > > Wow! That's a hell of a speaker! > > > Looking at your "ultrapiano" page, it is a load of nonsense. > No such thing has been built or ever will be built. > You don't have the knowledge of physics, acoustics, or > signals to know that it can't work as you describe it.
You seem to have found your way from the original message onto a page in a website unrelated to the subject under discussion. Did you click on something by accident and then become confused? At least you still have some long-term memory remaining - you can even remember your third-grade teacher! I'm sure she would have been proud of you. Old people often have difficulty with new technology, so you shouldn't fret or get anxious about it - at least you'll have something to tell the grandchildren next time they visit!
FWIW, I recently suffered some inner-ear damage due to a virus ?
specialists don't definitively know yet. At 1st I was completely deaf
in the left ear. Gradually the sound is coming back. I now get the
following properties in my perception of sound :-
Low freqs(<1000Hz) + High freqs(>3KHz) are ok ( well, as good as they
ever were)
Mostly silence from 1KHz -> 2.2KHz, but strong peaks of response at a
few spot frequencies.

The weidest thing is hearing the same pure sinewave a different
frequencies depending on whether I use the right ear or left ear. The
difference is like C to F# in the same octave.
The above figures are not exact, but are roughly what I get.

I had always imagined that perception of frequency would be invarient
from left to right ear, but now I know better.