ES-1 technical question re: sample rate

[AmigaHeretic]

I'll give an example using the ES-1 and a record player and a 7" 45rpm single as a demonstration, but it could also be a computer or a tape player or another synth/sampler, but the important thing is that it needs to be able to vary the playback speed of the sample, tape or record.

Put the 7" 45rpm single you want to sample on the record player, but set the record player to play at 33rpm - so that it plays the record slower and at lower pitch.

Well, my only thought on it would be this, I'll try to explain clearly...

Ok, I'll just make numbers up out of by butt as I really have no idea what I'm talking about, but don't worry about the numbers it's the idea...


Ok, so you are taking one sample at norml speed 45rpm and one sample at 33rpm.

So, lets say the sample you are talking is -- 10 seconds. ( at 45rpm) and lets also say a 10second samples takes up 1megabyte of space (again numbers made up.)

Now it seems to me if you are sampling at 33 rpm, them the sample is now going to take what 15seconds or something. So that mean you now have 1.5 megabytes of samples data.

45 rpm = 10 seconds = 1megabyte

33rpm = 15seconds = 1.5megabytes.


So looking at that it seems like you have quite a bit more sample information and in my mind that means higher quality.......

[fac]

45 rpm = 10 seconds = 1megabyte

33rpm = 15seconds = 1.5megabytes.


So looking at that it seems like you have quite a bit more sample information and in my mind that means higher quality.......

In this case, yes. However, once you take the 33 rpm sample (1.5 megabytes) and pitchshift it to its original tone (by digitally playing it at the equivalent of 45 rpm), it will be reduced again to 1.0 megabytes by means of interpolation or whatever other technique.

Basically, you're playing the original record at a slower speed to "capture" more information, but once in the digital domain, you're playing it faster, which in the digital domain means you skip a few samples and losing information, to have it at its original pitch.

So the question is whether this pitchshifted 33 rpm sample has better quality than the 45 rpm sample when they both occupy 1 Mb of space.

This is quite interesting, don't you think?

[gmeredith]

AmigaHeretic and Fac,

Thanks for discussing it that way - I think all the way along, I have been trying in my mind to get my head around the concept of "if it does work, why does it work".

I can see that Fac may be right in his explanation of the interplolation issue cancels out the higher resolution of memory, on the ES-1.

On my casio SK-8 lo-fi sampler, which spreads out a single sample across its 3 octave keyboard, this technique works quite dramatically, probably because the SK8 is a very early sampler which simply speeds up its sample playback as you play the higer notes (I think that's how it works on it), to achieve the musical scale. The ES-1, it seems, doesn't work that way, so something else is at work in the ES-1 to give the apparent improvememnt in quality.

I have an example of the principle recorded using my casio SK-8, if you want to judge the difference in the sound quality it makes of the SK8:

http://www.jz-server.de/forum2/e107_plu ... .php?614.0

Just for you guys - sample 1 on that page is the straight recorded sample, sample 2 is the pitch-up method sample. Let me know what you think.

Cheers, Graham

[n3ldan]

Let's pretend that you have a sampler with an adjustable playback and recording frequency, that can record 10 seconds at 32kHz.

10 seconds @ 32kHz means it has a total memory of 320,000 samples.
You can adjust the recording freq to 64kHz, which will be much higher quality, but you can now only record 5 seconds. 5*64,000=320,000 samples.

Now lets say you have a machine with a fixed recording sampling rate, but a variable playback sampling rate.

You record 10 seconds of a normal speed sample at 32kHz
You record 10 seconds of the same sample at half-speed, at 32kHz

You playback the first sample at 32kHz, lasting ten seconds. It sounds normal.
You playback the second sample at 32kHz, lasting ten seconds. It sounds like it was played at half speed.

You then playback the second sample, at 64khz, lasting 5 seconds. It sounds normal, but also much higher quality than the first sample.

Like you've said, it is doubtful that the esx or any other modern sampler does this. This is the old-school way to do it. This is how your lofi casio sampler does it, graham.

The esx has more information if you record a slower sample and then speed it up, but it still outputs at 32kHz. It is possible that an interpolation algorithm would change the sound but to what extent I do not know.

So the question is whether this pitchshifted 33 rpm sample has better quality than the 45 rpm sample when they both occupy 1 Mb of space.

It doesn't. They contain the same number of samples and thus are the same quality. They will sound different because of the algorithm that compressed 64,000 samples into 32,000 samples (assuming 1 second sample length).

[Borg]

A very, very interesting subject. The only thing missing...


WAVS!

gmeredith, post the two samples here from the ES-1, like you did with the ones from the SK-8. I think I'm not the only one curious on how it sounds like!

[gmeredith]

Hi Borg,

I haven't figured out how to post wavs up on this forum yet (other than just linking them to the site I mentioned). I'll put some ES-1 wav's up on that other site soon (my ES-1 been borrowed for week or 2 by a friend wanting to do the experiment also!).

In the meantime, I conducted the experiment on another sampler - a Casio FZ-1, at a sampling rate of 18k. The wav files of it have been added to the original page here:

http://www.jz-server.de/forum2/e107_plu ... p?614.last

In the FZ-1 example, the trick method sample is "fz sample 2.wav". The normally-sampled one is "fz sample 1.wav"



n3ldan:

That is a really good way to explain it, that makes lots of sense to me!
And the resulting sample time lengths conform exactly to your description. The trick method sample is exactly half the length of the normally sampled one, even though their pitch and tempo are the same, obviously packing higher memory resolution into the the same equivalent sample time. So the downside of this quality increasing method is that you only get half the sampling time normally available from your sampler.

Cheers, Graham

[plosive]

this same thread has been going on and on and on on the yahoo group and its rather annoying.

man, just do yourself a favor and learn about DSP basics and sampling/bit rates instead of taking guesses and saying everyone else is incorrect after they reply (which is the same thing as ignoring people after it is clearly explained). hell.. just spending 5 minutes at wikipedia would do you some good.

if you think something sounds better then do it, but it has nothing to do with the quality or tricking the sampler into using another sample or bit rate.

even in software such as a wave editor, if you use a different sampling rate to playback a sample at a higher speed than the one it was created in.. it is through interpolation.. the final output will still be the bitrate at which the playback engine is using. this is how aliasing happens in the first place. frequencies that are 'lost' above nyqust are sometimes miss handled, poorly filtered or inaccurately placed in the spectrum when the interpolation is applied. some dsp engines handle this better than others. there are alot of interpolation quality comparisons for different samplers in strong detail all over the internet. and FYI.. no matter what "method" is discussed regarding pitching the samples up or down.. it is all about DSP. DSP doesnt mean "effect". DSP is everything.

Dont take me wrong though. it's good that you're exploring things and trying to find out more and how things work.. but instead of running circles around several correct answers and advice, or getting offended when someone tells you that you are wrong (re: yahoo group) listen for a second and take a minute or two to brush up on what it is you are actually talking about.

And before you rant against me.. you're using the term "quality" itself inaccurately. Just because you or someone else might think something sounds better (aka relativity), this has no actual reference to the Technical (re: subject title) quality of the sampling rate, bit rate, dsp effect, or so on. It's simply about perception.

The simple fact of the matter is that a pitched sample, wether through an extremely complex and well done resampling technique or through some other form of pitchshifting (comb/delay/etc).. the end result is never going to be of better "quality" than the original. You may think it sounds better and it indeed might, but in reality you have less of the original sample than you had before and the quality is always on a diminishing scale. This is why ppl use romplers, or multisample, or use extremely expensive samplers such as a Roland's Variphrase or software solutions, etc., that focus as much of their attention as possible on resampling techniques and interpolation when they want as much high quality as they can get from sampling.

if you read this right, you should get that I'm not saying you're wrong, or they're right... im simply pointing out your misunderstanding regarding the technicals.

[gmeredith]

Hi plosive,

or getting offended when someone tells you that you are wrong

???? When did I sound offended by anyone's explanation over there?
Are you referring to Zoinky's comment? I just wanted him to listen to the wav files first before anything, that's all.

but instead of running circles around several correct answers and advice

You will note on that other forum that I said that I changed my mind and said that I no longer presumed to know the answer. Hence the idea to conduct the experiment.

Since this forum is a little bit behind in posting dates, I also state here that any theory I have previously expressed as to why this method does what it does, I now retract, and I now no longer presume to know how all of this works.

That leaves me with a bunch of recordings that sound different, resulting from 2 different ways of sampling.

What I am I am interested in is to find out if people have noticed this phenomenon also, and does it have an application in getting better sampling results out of samplers. What do you think?

The simple fact of the matter is ... the end result is never going to be of better "quality" than the original

"original" - you mean the original source sound, fed into the sampler?

No No No No No!!!!

That's NOT what I've been meaning all through out this. Is that what you and the others thought I was saying?
Is this what the issue on the other forum has been about all this time (other than the theory stuff/me not listening etc)??

I'm talking about the difference in results to each sampling method, NOT the difference to the original source sound.

I never said that this method will give a better sample than the original source sound. I said that it seems to give a better result compared to the result you get by the normal sampling way.

Cheers, Graham

[plosive]

hey its all good man, im not tryin to stir up the water. i think its good you're out there thinking about this. it is a good path i really do recommend reading up and i dont mean that as an insult. i personally am fascinated with dsp and have been for too many years now.. its my #2 hobby after synthesizers themselves

i will try out the technique you're talking about but i have a pretty good idea how it'll come out. i'm sure ive done it inadvertedly several times already actually.. however, i am getting ready to sell my es1 (own a esx now).

[gmeredith]

Hey, I think we keep editing our posts before each other reads them

I am in the process of reading some of the basic sampling theory stuff as we speak, especially some of the links that others gave. I'll see if I can understand it better.

Selling your ES-1 for an ESX?

Looking for 44k sampling - why don't you try my new sampling trick!! Just kidding

Cheers, Graham

[fac]

Looking for 44k sampling - why don't you try my new sampling trick!! Just kidding

I know you're kidding but I just want to make it clear: your trick, whether it works or not, would never be equivalent to sampling at a higher rate (in this case, 44.1 Khz). There's simply no way to regenerate the frequency content above 16 Khz that was lost due to sampling at 32 Khz. At least, not as it was in the original source.

I haven't been able to test your trick - I plan to do some tests in Octave, a numerical computation program, by additively sintesizing some signals with components over Nyquist frequency (> 16 khz), then sampling them at 32 Khz at normal and half speed, then pitchshifting the half-speed sample one octave higher using linear interpolation. I can tell you beforehand that, theoretically, the pitchshifted signal will have worse quality in the sense that it will be less alike the original signal, but I want to measure by how much.

[gmeredith]

Hi fac,

There's simply no way to regenerate the frequency content above 16 Khz that was lost due to sampling at 32 Khz. At least, not as it was in the original source

Totally agreed!

I think that I may have been misunderstood in this whole thing as to what I have actually meant. I just realised that I may have been using the wrong terminology while talking about it. I have been talking about sampling frequency, when I should have been talking instead of AUDIO BANDWIDTH.

Just quoting what I posted a few posts up:

"original" - you mean the original source sound, fed into the sampler?

No No No No No!!!!

That's NOT what I've been meaning all through out this..I'm talking about the difference in results to each sampling method, NOT the difference to the original source sound.

I never said that this method will give a better sample than the original source sound. I said that it seems to give a better result compared to the result you get by the normal sampling way.

So it's not about trying to get it better than the original - it's about comparing the results of 2 sampling methods, and seeing which gives better results compared to EACH OTHER - not the original

My original preposition of the trick relied on the fact that if you slow the incoming source sound down, it will no longer have high frequencies (ie. 22k audio bandwidth) to have to try and get recorded in the sampler. If you dropped it an octave, say, it would mean that the highest frequency in the incoming source sound is now 11k audio bandwidth. This is easily handled by the ES-1's 32k sampling rate, and is below the Nyquist theorem 16k audio bandwidth cutoff. Now, when you speed the sampled 11k audio bandwidith sound back up 1 octave on playback by playing it back, so I originally presumed, maybe then it plays those 11k audio frequencies now at 22k audio bandwidth (minus real-world losses by filters, etc - maybe 20k in reality). That was the thought I originally had to explain the apparent improvement in quality AS OPPOSED to sampling the 22k bandwidth source sound via the NORMAL sampling method - but either method is definitely NOT as good as the original 22k audio bandwidth sound recording itself.

As it is now, I don't know what mechanism causes the apparent increase in quality between the 2 SAMPLING METHODS.

My latest example using the FZ-1 sampler here seems to show a very obvious difference in quality, as heard by these recording examples (down the bottom of that page):

http://www.jz-server.de/forum2/e107_plu ... ic.php?614

I'm thinking that I should wind this topic up - it may be getting in the way of other topics people need to ask and is well past overstaying its welcome. Thanks heaps for everyone's input - you have all given me lots to think about and read up on!


Cheers, Graham

[fac]

Just quoting what I posted a few posts up:

"original" - you mean the original source sound, fed into the sampler?

No No No No No!!!!

That's NOT what I've been meaning all through out this..I'm talking about the difference in results to each sampling method, NOT the difference to the original source sound.

I never said that this method will give a better sample than the original source sound. I said that it seems to give a better result compared to the result you get by the normal sampling way.

My original preposition of the trick relied on the fact that if you slow the incoming source sound down, it will no longer have high frequencies (ie. 22k bandwidth) to have to try and get recorded in the sampler. If you dropped it an octave, say, it would mean that the highest frequency in the incoming source sound is now 11k bandwidth. This is easily handled by the ES-1's 32k rate, and is below the Nyquist theorem 16k cutoff. Now, when you speed the sampled 11k bandwidith sound back up 1 octave on playback by playing it back, so I originally presumed, maybe then it plays those 11k frequencies now at 22k. That was the though I originally had to explain the apparent improvement in quality AS OPPOSED to sampling the 22k bandwidth source sound via the normal sampling method - but either method is definitely NOT as good as the original 22k bandwidth sound recording itself.

As it is now, I don't know what mechanism causes the apparent increase in quality between the 2 SAMPLING METHODS.

Well, let's try to make things clear: one important thing is that there are not two sampling methods: there is just one - the one that the ES-1 uses, and in the ES-1 the output sample rate is fixed at 32 kHz, which means that, no matter what, you will never hear a frequency over 16 khz coming out of the ES-1.

Now, let's take your example: suppose you want to sample a sound with a 20 Khz component. If you sample it at 32 Khz, that component will alias and appear as a 12 Khz component in the digital signal. Now, suppose you slow down the original source to half its speed, so the 20 Khz component becomes 10 Khz. Now, if you sample the slowed-down signal at 32 Khz, you will effectively capture the 10 Khz component since it's well beyond Nyquist (16 khz). The problem is when you pitchshift the slowed-down sound up to achieve its original pitch. Pitchshifting will push the 10 Khz component back to 20 Khz, but since the sample rate is fixed at 32 Khz, that component will once again be reflected to 12 Khz.

I hope you can post wav files from the ES-1. I've listened to your casio samples, but since I have no idea how those samplers work, I cannot say that the same principle applies to them (especially if they've been modded).

[gmeredith]

I hope you can post wav files from the ES-1.

Coming soon!!

I've listened to your casio samples, but since I have no idea how those samplers work, I cannot say that the same principle applies to them

They're old school type samplers, 1980's, no effects as such, etc

Since starting the topic about the ES-1, I've learnt that the ES-1 does it differently to these, and so my thoughts about it won't be applicable to it, obviously. Like you say, it may be due to other factors.

Cheers, Graham

[fac]

They're old school type samplers, 1980's, no effects as such, etc

Yes, but this has nothing to do with effects. The principles of DSP apply to any sampled signal.

One possible explanation is that those casio samplers actually modulate the output sample rate for each voice in order to perform pitch-shifting. I think it's very unlikely, but this is the only way I can think of that could increase the output bandwidth. Some hybrid synths did it this way, but for other reasons (e.g., each voice had to be converted back to analog to go through an analog filter and VCA), but in a fully-digital synth it makes no sense to perform DA/AD conversions at the middle of the path. The FZ-1 came out in the late 80's so I very much doubt it had analog filters.