Linear Resampling for Winamp or Foobar

[Seidhepriest]

Well now.

Electronic music *requires* high sampling rates and high processing bit depth to sound natural. That is something anyone who's done any serious mixing in 24/32-bit will tell, also anyone who's heard a live synthesiser play and then the same performance played as a recording.

With even the best windowing-mode resampling algorithms, downsampling to CD frequency always makes a mix sound flatter, and lose much of its vividness during quantising to 16-bit (dithering only helps so much). Whichever theories and whatnot thrown in, there's the truth, and it's obvious that theories have to be designed based on current reality as well as the other way around (theories have to be capable of describing how a new creation will operate), but trying to "hammer" reality around a theory is a tad Inquisition-style, isn't it?

Gnome, that "18 KHz" limit is quite a lie. Here the hearing limit is 23.5 KHz (with the gear that's here, anyway). Sony lowpasses every consumer audio device at 18 KHz. Have you heard how compressed their gear sounds? Entire instrument parts, like tubular bells, synthesised pads, subtle strings, get blurred in the mix. They're barely audible. Compressing bandwidth may reduce the noise floor (which is what Sony is obviously trying to do), but it also removes clarity. Not very much point to it nowadays anyway, with really low noise floors at over 100 dB for professional recording equipment. CDA, by the way, is only 96 dB.

What's more, something people ignore completely, and just one of CDA problems, is the artificial clipping that happens to anything beyond 22050 Hz which distorts the CDA sound. There's actually a lot of sounds that go over 22 KHz (chromatic percussion, like hihats on a drumkit, usually does 30 KHz), in a CDA-frequency recording they're artificially cut, introducing "sawtooth" to the wave.

That many recent rock recordings have been maximised/compressed into the sky is also simple proof that with current recording and mastering equipment the CDA bandwidth isn't enough. Nevermind windowing lowpass filters - they're typically used to get rid of the CDA sampling aliasing which occurs as described above - out of not enough sensitivity. Also to remove noise by signal compression, but an artificially quiet recording is, of course, far worse than a more naturally sounding recording with noise.

Regardless, it's pretty much unclear what is the point of "defending" and imposing fanatism. There's a difference between blind reasoning and a discussion based on perceived reality, which often any reasoner theories bypass.

[LawnGnome]

Originally Posted by Seidhepriest

Well now.

Electronic music *requires* high sampling rates and high processing bit depth to sound natural. That is something anyone who's done any serious mixing in 24/32-bit will tell, also anyone who's heard a live synthesiser play and then the same performance played as a recording.

With even the best windowing-mode resampling algorithms, downsampling to CD frequency always makes a mix sound flatter, and lose much of its vividness during quantising to 16-bit (dithering only helps so much). Whichever theories and whatnot thrown in, there's the truth, and it's obvious that theories have to be designed based on current reality as well as the other way around (theories have to be capable of describing how a new creation will operate), but trying to "hammer" reality around a theory is a tad Inquisition-style, isn't it?

Gnome, that "18 KHz" limit is quite a lie. Here the hearing limit is 23.5 KHz (with the gear that's here, anyway). Sony lowpasses every consumer audio device at 18 KHz. Have you heard how compressed their gear sounds? Entire instrument parts, like tubular bells, synthesised pads, subtle strings, get blurred in the mix. They're barely audible. Compressing bandwidth may reduce the noise floor (which is what Sony is obviously trying to do), but it also removes clarity. Not very much point to it nowadays anyway, with really low noise floors at over 100 dB for professional recording equipment. CDA, by the way, is only 96 dB.

What's more, something people ignore completely, and just one of CDA problems, is the artificial clipping that happens to anything beyond 22050 Hz which distorts the CDA sound. There's actually a lot of sounds that go over 22 KHz (chromatic percussion, like hihats on a drumkit, usually does 30 KHz), in a CDA-frequency recording they're artificially cut, introducing "sawtooth" to the wave.

That many recent rock recordings have been maximised/compressed into the sky is also simple proof that with current recording and mastering equipment the CDA bandwidth isn't enough. Nevermind windowing lowpass filters - they're typically used to get rid of the CDA sampling aliasing which occurs as described above - out of not enough sensitivity. Also to remove noise by signal compression, but an artificially quiet recording is, of course, far worse than a more naturally sounding recording with noise.

Regardless, it's pretty much unclear what is the point of "defending" and imposing fanatism. There's a difference between blind reasoning and a discussion based on perceived reality, which often any reasoner theories bypass.

Your saying that the 18khz average max hearing for an adult is a lie? I'd disagree as many studies have shown otherwise.

[b0dhi]

Originally Posted by LawnGnome

Your entire thread is a threadcrap to this forum.

You post a thread asking for advice, and then ignore everything.

No, I posted asking for a linear resampler for Winamp or Foobar. Then you barge into the thread rallying about my "misconceptions". It's perfectly clear that you have no idea what you're talking about, since you did not even understand what the digitalproducer article I quoted was talking about. Stop being a troll.

Originally Posted by LawnGnome

Your saying that the 18khz average max hearing for an adult is a lie? I'd disagree as many studies have shown otherwise.

Yeah, in 6 paragraphs thats what he said *rolls eyes*. Oh, and learn to spell. Actually, learn to spell, learn something substantial about audio, then come back.

[Seidhepriest]

For starters, it's "you're". Trollspeak indeed...

Anyway, "agree" or "disagree" doesn't matter here, as in terms of digital sound processing, sampling frequency is just another part of overall resolution. For distortion-free transmission of recorded sound, at least triple the base frequency (22050 Hz in this case) is required, somewhere around 70 KHz. And 24-bit sample values to avoid distortion during processing (waves can be stored with 16-bit sample values, but they require 24-bit just to avoid noise introduced by rounding errors during transforms).

In this sense, 96000/24-bit is a decent (if only a bit "cold" to the sounding compared with analogue media) format as it both allows the high-frequency parts of a recording to be stored (therefore storing the sound in its wholeness, not cutting it forcibly like CDA), and removing the need to quantise down to 16-bit.

What the hypothetical *average* human (or some other mythical creature) hears consciously, doesn't matter at all if only because the frequency bandwidth determines how much detail can be accurately transmitted (as it has already been written above - the 22050 KHz recording rate of CD audio introduces distortion simply by discarding parts of sounds being recorded, whether a listener can notice anything above 22050 KHz is irrelevant as the frequency limit itself is the distorter).

96-KHz/24-bit audio is noticeably more clear and detailed than CD audio. Try the "Dark Side of the Moon" 30-th anniversary SACD to sample, both in CDA and 96-KHz mode.

[Hancoque]

Originally Posted by Seidhepriest

What the hypothetical *average* human (or some other mythical creature) hears consciously, doesn't matter at all if only because the frequency bandwidth determines how much detail can be accurately transmitted (as it has already been written above - the 22050 KHz recording rate of CD audio introduces distortion simply by discarding parts of sounds being recorded, whether a listener can notice anything above 22050 KHz is irrelevant as the frequency limit itself is the distorter).

How so? Normally there's a lowpass filter applied to the material before downsampling. If that lowpass filter cuts off frequencies above what most adult humans can hear (lets say 20 kHz) than you shouldn't notice anything because you wouldn't be able to hear what's missing in the first place.

I did a test not long ago. I tested which frequencies I was still able to hear. 17 kHz was no problem. 18 kHz was significantly quieter and 19 kHz seemed totally silent to me (could also be the headphones). Then I tested at which frequency cut-off in normal music I would hear a difference to the unfiltered material. This was significantly lower, at about 14 kHz. So, even if you notice some piercing in your ear at 20 kHz, this won't matter much in regards to music. I don't say that 14 kHz are enough for you but 20 kHz should really be. And I also doubt that music is produced consciously above that frequency. I rather deem this an uncontrolled rest of the recording process. It's like painting pictures with fractions of infrared or ultraviolet in it. Nobody needs that because nobody can see it. You would only annoy your dog with those high frequencies.

P.S.: We could do some ABX tests. That's an offer by me to avoid any "shut up, you don't know ****" replies. Just use a high quality music sample with a sample rate of 96 kHz and apply a high quality lowpass filter with different cut-off frequencies like 30 kHz, 24 kHz, 22 kHz, 20 kHz, 18 kHz and 16 kHz. ABX seems to be the only way to convince each other, I guess. Of course, cheating by looking at the frequency spectrums would be self-deception.

[LawnGnome]

Originally Posted by b0dhi

No, I posted asking for a linear resampler for Winamp or Foobar. Then you barge into the thread rallying about my "misconceptions". It's perfectly clear that you have no idea what you're talking about, since you did not even understand what the digitalproducer article I quoted was talking about. Stop being a troll.



Yeah, in 6 paragraphs thats what he said *rolls eyes*. Oh, and learn to spell. Actually, learn to spell, learn something substantial about audio, then come back.

You have yet to address anything. Most likely because you can't. So you resort to nit-picking my spelling.

But, Just to point out to you, it was a grammar problem, not a spelling problem. Not like it matters, since in this thread you have used "coz" and made several other spelling and grammatical errors.

Maybe you should focus your time on reading comprehension, instead of checking to see if I use the correct "you're/your", so that way you could actually understand the articles you link to.

But I think I'll let you get back to finding that linear sampler, so that you can attempt to pull "harmonics" above 22khz from a CD, since you think it will give you better sound quality.

[b0dhi]

Originally Posted by Hancoque

How so? Normally there's a lowpass filter applied to the material before downsampling. If that lowpass filter cuts off frequencies above what most adult humans can hear (lets say 20 kHz) than you shouldn't notice anything because you wouldn't be able to hear what's missing in the first place.

I did a test not long ago. I tested which frequencies I was still able to hear. 17 kHz was no problem. 18 kHz was significantly quieter and 19 kHz seemed totally silent to me (could also be the headphones). Then I tested at which frequency cut-off in normal music I would hear a difference to the unfiltered material. This was significantly lower, at about 14 kHz. So, even if you notice some piercing in your ear at 20 kHz, this won't matter much in regards to music. I don't say that 14 kHz are enough for you but 20 kHz should really be. And I also doubt that music is produced consciously above that frequency. I rather deem this an uncontrolled rest of the recording process. It's like painting pictures with fractions of infrared or ultraviolet in it. Nobody needs that because nobody can see it. You would only annoy your dog with those high frequencies.

P.S.: We could do some ABX tests. That's an offer by me to avoid any "shut up, you don't know ****" replies. Just use a high quality music sample with a sample rate of 96 kHz and apply a high quality lowpass filter with different cut-off frequencies like 30 kHz, 24 kHz, 22 kHz, 20 kHz, 18 kHz and 16 kHz. ABX seems to be the only way to convince each other, I guess. Of course, cheating by looking at the frequency spectrums would be self-deception.

You need to move beyond "the limit of hearing is 20Khz. sampling frequencies higher than that is needless" kind of reasoning because it's overly simplistic. As the digitalproducer article I quoted points out, phase information is important for binaural hearing (ie. soundstage, imaging etc), particularly inter-aural time delay, which, in the average individual, resolves better than 48Khz can sample (less than 15microseconds). Some people can resolve ITDs of less than 5microseconds, which requires over 192Khz sampling rate to resolve. We're not talking in the frequency domain here, we're talking about phase relationships, which the ear/ears can pick up. So the higher sampling rates aren't necessarily so that we can imagine hearing 200Khz harmonics, it's so that we can tell precisely when signals we can hear, arrive.

All this is very clearly explained in the digitalproducer article.

[Hancoque]

Okay, I have read the article but I still don't get it because you aren't confronted with a series of single samples while listening (like a sample machine gun). That's how it is represented in the digital domain, but analog converters should smooth these out, so that it looks exactly like the pictures I showed you before.

It’s been determined that time delay differences of 15 microseconds between left and right ears are easily discernible by nearly anyone.

What does that exactly mean? Why between left and right ears? If I would send a single sample to the head, there would be a delay between the arrival of the sonic wave at the left ear and right ear if the signal doesn't come from a centered source. But that would always be the case, regardless how quick the samples come after another. I really don't understand what the article means. Would samples with higher sample rates be equally effective if they wouldn't contain any more high frequency information than samples with lower sample rates? Or is it tied to that?

Update:
I was looking for comments on the article to see how the feedback is and there really is one from 2006:

I'm sure nobody is still reading this, but I just came across it. I had to comment, because it's clear the author is misinformed about sampling theory. It is true that people can discern delays less than the sampling period of a CD. However, that doesn't mean we're not sampling enough. The delay precision that can be recovered from a sampled signal is a function of the quantization, not the sampling rate! In theory (with infinite number of bits per sample) there is no limit to the precision with which a delay (phase) can be reproduced.

That makes some sense to me.

[LawnGnome]

Originally Posted by Hancoque

Okay, I have read the article but I still don't get it because you aren't confronted with a series of single samples while listening (like a sample machine gun). That's how it is represented in the digital domain, but analog converters should smooth these out, so that it looks exactly like the pictures I showed you before.

What does that exactly mean? Why between left and right ears? If I would send a single sample to the head, there would be a delay between the arrival of the sonic wave at the left ear and right ear if the signal doesn't come from a centered source. But that would always be the case, regardless how quick the samples come after another. I really don't understand what the article means. Would samples with higher sample rates be equally effective if they wouldn't contain any more high frequency information than samples with lower sample rates? Or is it tied to that?

Update:
I was looking for comments on the article to see how the feedback is and there really is one from 2006:
That makes some sense to me.

Exactly what I was trying to point out to b0dhi, that the author does not understand what he is talking about.

Maybe b0dhi will listen now that I'm not the only one saying, but I'm pretty sure he will just go on with his illogical ideas based on articles where the author is very misinformed.

EDIT:, Han, the thing about the 15ms is if 2 sources are playing the same sound, and one is delayed by around 15ms, it will sound as if there is only one source playing the sound. Only the earlier source will be heard. but, this does not apply to the sample rate, and I'm not sure why the author would bring it up, as it does not relate to his argument. So as for why he mentions it, I am just as confused as you.

[b0dhi]

Hancoque, regarding the comment you quoted, in theory it's correct. In practice, we don't have infinite bit-depth, and we don't have ideal reconstruction algorithms (or at least we don't have infinite processing power). Here's a quote from Ryohei Kusunoki, someone with considerably more authority than someone posting a comment on the internet: -

So, what is the sampling frequency in essence? Sampling the sound with 44.1kHz means that the CD can "differentiate the sound up to 25 microseconds." Raising the sampling frequency to 96kHz, for example, should not be considered as an extended frequency range up to 48kHz; it should be regarded as an "enhanced precision - over time domain," instead.

From http://www.tnt-audio.com/intervis/kusunoki_e.html.

Also, regardless of what anyone says, I know I can hear a difference between an unfiltered square wave and one filtered at 22050Hz, even though - theoretically - I shouldn't.