Dear all,
more for experimenting than of any practical value I would like to ask
this nice community for inputs & opinions on two related, highly theoretical projects,
the first one realistic, the second one utopia. I asked ChatGPT a bit for answers, but I'm not sure
how accurate the information gained is.
Project #1 (software):
Task: Retrieve digital audio samples of very high quality.
Split the audio range (20 Hz - 20000 Hz) of these samples into non-overlapping, logarithmically spaced bands of a variable number "n".
"n" variable for experimenting, f.ex. n = 1 up to 1000.
Measure total amplitude of every single band and apply this amplitude to a sinus wave of each
band's center frequency. Add all these new sinus waves together again to receive a summed signal which is a 'reduction'
of the source's audio signal (Fourier etc.).
Base Questions: How many bands "n" at a minimum are necessary to understand voice signals again from this source signal reduction
& resynthesis ? And how many at a minimum for 'rather' uncompromised audio quality ?
(Interestingly modern cochlear implants (of course no HiFi quality) use the very low number of 8 - 22 different frequency bands,
that are overlapping, however, and thus correspond possibly to 6 - 12 different bands only, effectively !)
Which software(s) then could do this job for sample source signals efficiently and accurately, but not necessarily in real time ?
(the problem is of course more complex, because not only amplitudes but also the phases of the signals also are important
across these manipulations).
Project #2 (practical utopia):
To build a loudspeaker with a large multitude of single low cost chassis.
Each chassis emits only one single fixed frequency, and each is triggered by an upstream
(cheap & as small as possible) amplifier that receives modulating input only from single sinus waves
(from Project #1).
Question: What is the minimum number "n" of chassis with logarithmically spaced fixed sinus signals
to recover 'rather' optimum audio quality (20, or 40, or 200, or 1000) ?
thank you kindly for any input & critique,
good night,
FrankieS
more for experimenting than of any practical value I would like to ask
this nice community for inputs & opinions on two related, highly theoretical projects,
the first one realistic, the second one utopia. I asked ChatGPT a bit for answers, but I'm not sure
how accurate the information gained is.
Project #1 (software):
Task: Retrieve digital audio samples of very high quality.
Split the audio range (20 Hz - 20000 Hz) of these samples into non-overlapping, logarithmically spaced bands of a variable number "n".
"n" variable for experimenting, f.ex. n = 1 up to 1000.
Measure total amplitude of every single band and apply this amplitude to a sinus wave of each
band's center frequency. Add all these new sinus waves together again to receive a summed signal which is a 'reduction'
of the source's audio signal (Fourier etc.).
Base Questions: How many bands "n" at a minimum are necessary to understand voice signals again from this source signal reduction
& resynthesis ? And how many at a minimum for 'rather' uncompromised audio quality ?
(Interestingly modern cochlear implants (of course no HiFi quality) use the very low number of 8 - 22 different frequency bands,
that are overlapping, however, and thus correspond possibly to 6 - 12 different bands only, effectively !)
Which software(s) then could do this job for sample source signals efficiently and accurately, but not necessarily in real time ?
(the problem is of course more complex, because not only amplitudes but also the phases of the signals also are important
across these manipulations).
Project #2 (practical utopia):
To build a loudspeaker with a large multitude of single low cost chassis.
Each chassis emits only one single fixed frequency, and each is triggered by an upstream
(cheap & as small as possible) amplifier that receives modulating input only from single sinus waves
(from Project #1).
Question: What is the minimum number "n" of chassis with logarithmically spaced fixed sinus signals
to recover 'rather' optimum audio quality (20, or 40, or 200, or 1000) ?
thank you kindly for any input & critique,
good night,
FrankieS
I'm aware of course of the recent thread "Fourier Transform Speaker" in diyAudio.
But I am more interested in the theoretical Fourier separation and resynthesis part to
understand how many driver chassis would really be necessary.
But I am more interested in the theoretical Fourier separation and resynthesis part to
understand how many driver chassis would really be necessary.
If we look on which notes you want to produce (music) you will find a minimum amount of drivers to cover all 12 notes in an octave and you have 8 octaves (96 notes) .
Then i guess there is information from ambience subharmonics, harmonics, and sound from fingers striking the string etc. So minimum 100 for sound and 300 for HiFi?
Then i guess there is information from ambience subharmonics, harmonics, and sound from fingers striking the string etc. So minimum 100 for sound and 300 for HiFi?
For an exact reproduction of the original signal, the project is impracticable. Consider a single mono audio track say 200 s long with data from a CD with a sample rate of 44.1 kHz. This has 8,820,000 samples. If you now perform the Fourier transform, then there will be 4,410,000 complex (amplitude + phase) components in the frequency domain. So in principle to reproduce the original track you would require nearly 4.5 million independently adjustable sine sources to reproduce the signal.
You could consider a compromise - how much detail can you discard without a listener detecting the loss? This is exactly the problem addressed by mp3 compression algorithms. Depending on the chosen bit rate, the amount of data can be reduced by a factor of 3 to maybe 20. For high quality audio with complex music (saxophone is notoriously difficult to reproduce I believe) you will need to the highest bit rates - skilled listeners can sometimes hear faults with 256 kb/s data, but almost no-one can hear faults at 320 kb/s. But even with a factor of 20 reduction, you'd still require half a million sources. I don't know exactly how the MP3 algorithms work, but I expect it will be very difficult to translate the code to the programming for a finite number of fixed sources.
Another compromise, is to shorten the time period over which you reproduce the signal - lets say 0.1 sec. Then you achieve a 2000 x reduction in the number of sources, but you must update the programming of > 2,000 sources every 0.1 sec without glitches.
It might be easier to hire the musicians.
You could consider a compromise - how much detail can you discard without a listener detecting the loss? This is exactly the problem addressed by mp3 compression algorithms. Depending on the chosen bit rate, the amount of data can be reduced by a factor of 3 to maybe 20. For high quality audio with complex music (saxophone is notoriously difficult to reproduce I believe) you will need to the highest bit rates - skilled listeners can sometimes hear faults with 256 kb/s data, but almost no-one can hear faults at 320 kb/s. But even with a factor of 20 reduction, you'd still require half a million sources. I don't know exactly how the MP3 algorithms work, but I expect it will be very difficult to translate the code to the programming for a finite number of fixed sources.
Another compromise, is to shorten the time period over which you reproduce the signal - lets say 0.1 sec. Then you achieve a 2000 x reduction in the number of sources, but you must update the programming of > 2,000 sources every 0.1 sec without glitches.
It might be easier to hire the musicians.
thank You, ESL 63, for your answer.
I guess, it is not only the 96 note basis, if You think f ex on complex synthesizer sounds and everything else.
I asked ChatGPT before and AI suggested to use MATLAB or OCTAVE software to reduce the sources sounds
and then mix them again. I feel it would be much fun to listen to the reduced versions of human voice and
also to find out which logarithmic division of 20 (or 16) to 20000 Hz would be minimum for acceptable
audio sound.
I guess, it is not only the 96 note basis, if You think f ex on complex synthesizer sounds and everything else.
I asked ChatGPT before and AI suggested to use MATLAB or OCTAVE software to reduce the sources sounds
and then mix them again. I feel it would be much fun to listen to the reduced versions of human voice and
also to find out which logarithmic division of 20 (or 16) to 20000 Hz would be minimum for acceptable
audio sound.
thank You, dear golfnut, for your valuable input. yes, surely with Fourier transform, reduction plus resynthesis quality will be suffering.
surely the MP3 developers have done a lot of scientific quality test sessions with experienced listeners.
surely the MP3 developers have done a lot of scientific quality test sessions with experienced listeners.
My estimation was based on instruments that is in tune. And in one tuning ex 440Hz, but if someone tunes the song in 438... then your speaker will be quiet!
Depending on the Q value of the notch filters you have to use to filter out each individual frequency...
I recently listened to a 4" single driver speaker with no filters at all. (I have had similar speaker myself before)
Tube amp driven. Wow...
Depending on what you desire from your listening experience... but they do something with the room, perspective, 3D... earthquake bass? No... 🙂
Depending on the Q value of the notch filters you have to use to filter out each individual frequency...
I recently listened to a 4" single driver speaker with no filters at all. (I have had similar speaker myself before)
Tube amp driven. Wow...
Depending on what you desire from your listening experience... but they do something with the room, perspective, 3D... earthquake bass? No... 🙂
So now with @esl 63 and @golfnut inputs we have 4,410,000 very small oscillators reproducing a fraction of the sound or 8,820,000 for stereo, or 22,050,000 for five channel surround. What if we printed the little oscillators onto panels and covered the walls with billions and billions of little oscillators?
sure, many things unthinkable are easily possible 100 years later.
by the way, perfect concept: wall paper loudspeaker, factually millions of small micro or mini chassis
with electronically adjustable directions like RADAR phased arrays ... somehow ...
by the way, perfect concept: wall paper loudspeaker, factually millions of small micro or mini chassis
with electronically adjustable directions like RADAR phased arrays ... somehow ...
Exactly what I was thinking. I saw a demo of a 'phased array' loudspeaker. It was made of 32 mini speakers mounted on a couple of sheets of plywood. The sounds were recorded by an array of 32 cheep microphones. The played back sound was pretty crappy frequency wise, but scary accurate as far as sounds in the bandwidth of the loudspeakers like birds or kids talking. It was kind of spooky.