Griffinator Posted August 10, 2003 Share Posted August 10, 2003 The reason "throwing out every other sample" does not and can not create a transparent sample rate change is twofold, and very simple. 1) A Sample Rate Converter does not check to see if the source is double (or half, as the case may be) the sample rate of the destination. It reconstructs and resamples the waveform. That's just how it works. There is a growing contingency of engineers who are of the opinion that exact multiple samplerates do create a more transparent conversion than non-multiples, but that does not mean that going 88.2 to 44.1 is completely transparent - only that it is more so than going 96 to 44.1 or 192 to 44.1. 2) In the process of converting from, for example, 88.2K to 44.1K, the SRC has to run new LP filters and new comb filters in order to remove the excess data (in the high frequency spectrums). This process in and of itself affects the final product. There is no such thing as a completely transparent high-pass filter. Again, I insist, that resampling a file to exactly half the prior samplerate involves more than just removing every other sample - there's a lot more to it, and those extra steps are what affect the quality of the output. It may be less damaging by subtle degrees than resampling from a non-multiple, but it is still damaging, and it is not a mere dropping of samples. Also, you do more harm by truncating a file from 24 to 16 than by dithering - unless your noise floor prior to truncation is higher than -96dB. You've already indicated that your equipment is reasonably good quality - it takes some really cheap gear to generate that high a noise floor at 24 bit wordlength. Quote Link to comment Share on other sites More sharing options...
Marvel Posted August 10, 2003 Share Posted August 10, 2003 Andy, It is amazing to me that digital audio works at all. Think of this example, using a picket fence for illustration. This is very simplified. Each picket represents one sample of a waveform, in actuality, the amplitude of that waveform when the sample is taken. All the pickets end up being different heights, because the amplitude is always changing. Using Dave's idea of more dots shows this very well also. If you connect the tops of all the pickets (using a straight line from top to top), it can be pretty jerky. Now add more pickets for the same time period, and connect the dots. It begins to look a lot smoother. The funny thing then is that playing back all of those samples (pickets) at the proper amplitude repoduces the sound of what was recorded. The Nyquist theorem, says you need double the samples for the frequency you want to reproduce. Since the audio world has used 20-20k as the audio spectrum, a value just over 40k was chosen. This was to have filters that would quickly chop off anything above the 20k, to not mess up the ADC. I think that Griff is saying that more samples helps (we all pretty much agree on that I think), but that you can still filter out everything above the 20K, and a smoother filter going to almost 24K is easier to make than one of a lower value with a sharper cutoff. (Griff, you can correct me if I've misunderstood you on this) You can use motion picture film as an example as well. Standard frame rate is only 24 fps. When those are played back in sequence, we see smooth motion. Until you have something moving very fast, and the movement between each frame becomes very large and it sometimes doesn't look as smooth as we would want. If you shoot 60 frames per second, and play back at the same rate, it looks very smooth indeed. I know this is a simplistic look at digital audio. This leaves out error corrections, L/R channels from a single data stream, and soooo much more. Your example of converting a DAT to 44.1 is one valid use of a SRC. Digital video (mini-DV and DVCam) use the 48K sample rate as well. Unless the audio is recorded separately the same way it is done in film production, you would have to deal with a 48K rate. Almost all audio is redone anyway (dialog, rustling papers and fabrics, traffic noise, etc.), so the original is of little consequence for film production. Marvel Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 10, 2003 Author Share Posted August 10, 2003 Arggh. Now I understand why we are missing each other. I do NOT resample to get from even multipliers like 88.2 to 44.1 One just resets the sample rate WITHOUT resampling. This simply disposes of now redundant data. Dave Quote Link to comment Share on other sites More sharing options...
AndyKubicki Posted August 11, 2003 Share Posted August 11, 2003 Marvel, believe it or not, I find it more impossible that a needle can track a groove without jumping out while tracking frequencies from 20 Hz (and lower) to 20 KHz. (I know, that's why there is an RIAA curve, but still...) Those are some phenominal forces! This presents another question...is there signal content in vinyl above 20 KHz? I believe it was filtered out. A lot of cartriges don't reproduce that high anyway, though if there is such an affect, it might take place without the cartridge actually reproducing those frequencies. If this is true, then perhaps we are barking up the wrong tree as far as the effect of ultrasonic frequencies on the audible range. As to the Nyquist theory, let me venture into shaky grounds here for myself. If 44.1 KHz is chosen because it's twice the upper end of what some humans can hear (22.05 KHz), he is only looking at a snapshot of the peakof that (highest freq) wave and the next digital snapshot is at 1/2 the wavelength. Aren't higher frequencies getting shortchanged here? If so, why do CDs sound brighter? Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/10/2003 4:27:45 PM Mallett wrote: Arggh. Now I understand why we are missing each other. I do NOT resample to get from even multipliers like 88.2 to 44.1 One just resets the sample rate WITHOUT resampling. This simply disposes of now redundant data. Dave ---------------- No, Dave, what you're misunderstanding is this: While you believe that you are just "resetting the sample rate", your software is resampling the data, whether you are aware of it or not. Doesn't matter what sample rate you go to or from, it quickly reconstructs and resamples the data. Sure, I've got a little radio button panel in WaveLab under "convert sample rate" - and from my perspective as the interfacing user it's just a matter of picking a sample rate and hitting OK. Meanwhile, in that brief amount of time where it says "converting sample rate" with a % done bar, it's doing exactly what I have described. Reconstructing and resampling. ----------------- I think that Griff is saying that more samples helps (we all pretty much agree on that I think), but that you can still filter out everything above the 20K, and a smoother filter going to almost 24K is easier to make than one of a lower value with a sharper cutoff. (Griff, you can correct me if I've misunderstood you on this) ----------------- I'm saying that 48K helps, and I'm saying that 96K helps, but that anything above 96K should be completely unnecessary for a 44.1K final product. The only reason to do 192K is if you're authoring a 24/96 DVD - for the reasons I outlined previously - making sure that effect tails, crossfades, etc are all handled by the dithering algo and are pushed outside of the frequency spectrum of the final product. I don't believe that tracking, mixing, and/or mastering at 192Ks/S will get you a better Redbook product. You may hear the difference when you're working with it directly, but I guarantee that you won't hear a difference between 96K-->44.1 and 192K-->44.1. Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 11, 2003 Author Share Posted August 11, 2003 No Griff, my little radio button says "Reset Sample Rate-DO NOT RESAMPLE." I'll take a look at Wavelab sometime today. I don't use it much except for it's anayltical functions. If there is no way to reset without resample, I'd get myself some software that can. By the way, one way to confirm this is that it takes no time. Resampling a large file takes a while even on a fast computer. Resetting the sample rate is instantaneous. I've had to look at the properties before in order to convince myself it had been done. It is simply not necessary to resample even divisibles. Do the math. This is an established principle in diital audio production. As to your second point in the last response, except that the accepted higher sample rates for Redbook targets amongst those of us of the "avoid resampling at all costs" crowd are 88.2 and 176.4. Dave Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 11, 2003 Author Share Posted August 11, 2003 You are correct about WaveLab. Makes me scratch my head, as this is otherwise a superb package. Check out Sound Forge/Process/Resample If you check "Do Not Resample" and do 44.1>96, it happens instantly and you get a mess due to the math. 44.1>88.2 or 176.4, also instant and sounds, well, just like it did except it is now one of those rates. Same thing going down. I've been doing this for several years for CD output in order both to achieve a better sound for my own use, as well as the benefits of any editing or processing (very rare, but it happens) at the higher rate. For bit depth conversion, I apply highpass triangular and equal loudness contour. My assumption on the noise shaping is that my recordings have almose no noise anyway. If I ever get a golden eared young audiophile that tells me it's audible, I'd probably change to high pass. Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/11/2003 8:17:12 AM Mallett wrote: No Griff, my little radio button says "Reset Sample Rate-DO NOT RESAMPLE." I'll take a look at Wavelab sometime today. I don't use it much except for it's anayltical functions. If there is no way to reset without resample, I'd get myself some software that can. By the way, one way to confirm this is that it takes no time. Resampling a large file takes a while even on a fast computer. Resetting the sample rate is instantaneous. I've had to look at the properties before in order to convince myself it had been done. It is simply not necessary to resample even divisibles. Do the math. This is an established principle in diital audio production. As to your second point in the last response, except that the accepted higher sample rates for Redbook targets amongst those of us of the "avoid resampling at all costs" crowd are 88.2 and 176.4. Dave ---------------- I am duly corrected re: resampling. I was not aware that SF did the elimination routine on even divisibles. That really is the first I've heard of such a thing. Most SRC's don't check for even divisibles when they convert rates. As you pointed out, that is an opt-in for SF, not an automatic. Re: resampling. Actually, on my little 1.77Ghz Athlon XP it takes about 10 seconds for Wavelab to resample 48Khz-->44.1Khz. Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 11, 2003 Author Share Posted August 11, 2003 Well, if I were half intelligent I wouldn't have taken so long to figure out our miscommunication here. I am still amazed that WaveLab, CoolEdit, N-Track, do not provide this basic function. Anyway, kept my brain going, even if in the wrong direction Thanks for your participation. You have a lot to offer. Dave Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/11/2003 11:25:17 AM Mallett wrote: You are correct about WaveLab. Makes me scratch my head, as this is otherwise a superb package. Check out Sound Forge/Process/Resample If you check "Do Not Resample" and do 44.1>96, it happens instantly and you get a mess due to the math. 44.1>88.2 or 176.4, also instant and sounds, well, just like it did except it is now one of those rates. Same thing going down. I've been doing this for several years for CD output in order both to achieve a better sound for my own use, as well as the benefits of any editing or processing (very rare, but it happens) at the higher rate. For bit depth conversion, I apply highpass triangular and equal loudness contour. My assumption on the noise shaping is that my recordings have almose no noise anyway. If I ever get a golden eared young audiophile that tells me it's audible, I'd probably change to high pass. ---------------- When you upsample, nothing changes. It's still the same signal, with lots more sample points. Only purpose for this would be applying new effects, etc. and changing the waveform prior to dropping back down to 44.1. Only time you hear any kind of difference is downsampling. I still have to sit back and wonder what effect the ultrasonics have on the sampling snapshots. The way I understand it, if ultrasonic frequencies (>22.05Khz) exist in your CD, it is not a legal redbook CD. I could be mistaken about that, but if anything ultrasonic content would have a negative affect on the consumer decks ability to read the samples properly (overtones fouling up the reconstruction filter). Obviously, I'm groping there, and into territory I don't 100% understand. Noise shaping is part of what I was referring to - dither creates an artificial noise floor via a randomization sequence in order to mask the quantization errors created by truncation. Noise shaping creates a certain "analog" character to that artificial floor - instead of white noise (true random content) it colors that noise to make it sound more natural (since white noise does not exist in naturally). Noise shaping algorhithms were crucial to getting CD over the hump from its dismal beginnings. Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/11/2003 1:42:36 AM AndyKubicki wrote: As to the Nyquist theory, let me venture into shaky grounds here for myself. If 44.1 KHz is chosen because it's twice the upper end of what some humans can hear (22.05 KHz), he is only looking at a snapshot of the peakof that (highest freq) wave and the next digital snapshot is at 1/2 the wavelength. Aren't higher frequencies getting shortchanged here? If so, why do CDs sound brighter? ---------------- The nyquist theory states that for one to correctly represent frequency N, one must sample 2N times per second. The Redbook standard demands a lowpass filter to eliminate supersonic frequencies that would not be correctly represented by the 44.1Ks/S (please don't say Khz - that's a frequency, not a samplerate.) Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 11, 2003 Author Share Posted August 11, 2003 >I've been doing this for several years for CD output in order both to achieve a better sound for my own use, as well as the benefits of any editing or processing (very rare, but it happens) at the higher rate. Bad choice of words on my part. What I meant was I location record at 24/88.2 in order to get better sound on my system but be able to drop to Redbook with little or no audible loss. Which do your prefer, High pass or equal loudness contour for noise shaping? As I mentioned, my high frequency hearing is not so hot. I use equal loudness and no ones ever mentioned it, so I assume it is OK. The only basis for this decision is my knowledge that there are those out there who can actually hear at the Redbook Nyquist point, and I don't want a bunch of noise there for them. Rather have it dispersed. Don't know whether that is good logic or not. What do you think? Dave Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 Well, all I can tell you about your choice of noise shaping algos is this: I have demonstrated (6 years ago at a pre-employment physical) a hearing range upwards of 24Khz. I have yet to hear a significant difference between the noise algos out there, but I can definitely hear the difference between a dithered wave and a truncated one. It's subtle - you really can only hear it when the signal approaches the noise floor (in a reverb tail or a fade) - because that's its fundamental job - to mask rounding errors at the noise floor (-96dB in 16bit, -144dB in 24bit) Quote Link to comment Share on other sites More sharing options...
Mallette Posted August 11, 2003 Author Share Posted August 11, 2003 Thank you. Just what I wanted to know. Wish I had my extended high freq back at least long enough to see what the 22.5 khz barrier sounds like. I never knew just how far out my teen/20's hearing went, but those "ultrasonic" horns used in motion detectors with their 140db SPL were brutal when I unknowningly walked under one. I was told they just switched off the alarm and left them going...since no one can hear them! I also threw away two "ultrasonic" dog whistles, because I could hear them so clearly I thought they were either BS or messed up. Well, that was a long time ago... Dave Quote Link to comment Share on other sites More sharing options...
AndyKubicki Posted August 11, 2003 Share Posted August 11, 2003 The nyquist theory states that for one to correctly represent frequency N, one must sample 2N times per second. The Redbook standard demands a lowpass filter to eliminate supersonic frequencies that would not be correctly represented by the 44.1Ks/S (please don't say Khz - that's a frequency, not a samplerate.) Forgive me if I ask a basic question here...when you say lowpass filter, I understand it as a filter that eliminates everything above 22.05 KHz, right? That still does not explain to me why a Redbook CD sounds bright compared to vinyl...to the point that I had thought the mastering process of Redbook CDs was where the brightness originated. And not only brightness, but lead instruments and vocals seem more "in your face" on Redbook than vinyl. I understand what you're saying about Nyquist, but if more dots make for a clearer picture, then the upper part of the audible spectrum is where anyfaults with the theory would show up. If doubling that sampling rate (or multiplying it) makes the recording sound better, then obviously is there not a flaw with that theory? Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/11/2003 3:26:43 PM AndyKubicki wrote: The nyquist theory states that for one to correctly represent frequency N, one must sample 2N times per second. The Redbook standard demands a lowpass filter to eliminate supersonic frequencies that would not be correctly represented by the 44.1Ks/S (please don't say Khz - that's a frequency, not a samplerate.) Forgive me if I ask a basic question here...when you say lowpass filter, I understand it as a filter that eliminates everything above 22.05 KHz, right? That still does not explain to me why a Redbook CD sounds bright compared to vinyl...to the point that I had thought the mastering process of Redbook CDs was where the brightness originated. And not only brightness, but lead instruments and vocals seem more "in your face" on Redbook than vinyl. I understand what you're saying about Nyquist, but if more dots make for a clearer picture, then the upper part of the audible spectrum is where anyfaults with the theory would show up. If doubling that sampling rate (or multiplying it) makes the recording sound better, then obviously is there not a flaw with that theory? ---------------- Ahhh. Let's clarify our terms for a moment. "Brightness", as you are understanding it, is not a product of extremely high frequencies. In fact, it is typically the 10-12Khz range of the spectrum that gives us this sense. It's how Bo$e makes their speakers sound so deceptively "clear", even though they are totally inaccurate. Boosting the "presence" or "brightness", i.e. 10-12Khz, is a common trick in country and pop music production. The other thing I have observed is that vinyl that has been used and abused over the years has a very dark quality, i.e. lacking in HF detail - probably due to worn grooves. RE flaw in the Nyquist theory: please re-read the thread and you will understand my position about increased sampling frequencies. Oh - and you are correct. A lowpass filter is one that allows everything below it to pass, while blocking everything above it. Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 ---------------- On 8/11/2003 3:09:26 PM Mallett wrote: Thank you. Just what I wanted to know. Wish I had my extended high freq back at least long enough to see what the 22.5 khz barrier sounds like. I never knew just how far out my teen/20's hearing went, but those "ultrasonic" horns used in motion detectors with their 140db SPL were brutal when I unknowningly walked under one. I was told they just switched off the alarm and left them going...since no one can hear them! I also threw away two "ultrasonic" dog whistles, because I could hear them so clearly I thought they were either BS or messed up. Well, that was a long time ago... Dave ---------------- Indeed, I know exactly the sensors you're referring to. Nasty sinewaves from out of the corners of the rooms. When I was a kid I could hear those things clear as a bell. Haven't been around any lately, couldn't tell you if I could hear that sound now... Quote Link to comment Share on other sites More sharing options...
AndyKubicki Posted August 11, 2003 Share Posted August 11, 2003 Ahhh. Let's clarify our terms for a moment. "Brightness", as you are understanding it, is not a product of extremely high frequencies. In fact, it is typically the 10-12Khz range of the spectrum that gives us this sense. It's how Bo$e makes their speakers sound so deceptively "clear", even though they are totally inaccurate. Boosting the "presence" or "brightness", i.e. 10-12Khz, is a common trick in country and pop music production. The other thing I have observed is that vinyl that has been used and abused over the years has a very dark quality, i.e. lacking in HF detail - probably due to worn grooves. You are right, it's not at the top of the freq range, but it is close. We never hear complaints of poor bass on Redbook. As to your observation about older vinyl, I would suspect that any damaged grooves would produce distortion, I never thought about the HF detail being worn out without noticable distortion, but that's just my experience. We all know about the natural loss of HF material as the groove nears the center. As to my own collection, I have tried to take good care of my vinyl back when I did play it, though my older ones have more wear than the newer ones. I was comparing (specifically) a fairly unplayed copy of Alan Parson's Project I Robot (I believe it was a half speed master) to the CD version. If what you are saying about the presence bump in the 10-12KHz range happened here, then my initial observation was correct...it's not the meduim itself, it was mastered that way. How long have they been using this "trick"? This can explain exactly what I'm hearing. BTW, I do have what I consider to be excellent sounding commercial Redbook CDs. One is the Rite of Strings with Stanley Clark on acoustic bass, Al Di Meola on acoustic and some electric guitar and Jean Luc Ponte...mostly acoustic and it sounds GREAT. Yet I wonder, how would this have sounded on Dave's 24/196...or SACD, DVD-A or any other hi-rez player... Our ears get accustomed to our listening environment and everything is releative there... Quote Link to comment Share on other sites More sharing options...
Griffinator Posted August 11, 2003 Share Posted August 11, 2003 Really, they've been doing it since CD was a new medium - the labels demanded it from the mastering houses - they wanted people to perceive CD's as sounding "clearer" than vinyl or tape - so the mastering houses responded by boosting presence. Quote Link to comment Share on other sites More sharing options...
AndyKubicki Posted August 11, 2003 Share Posted August 11, 2003 Thank you Griff!!! That is what my ears have been telling me! Another paradigm shift, back now to where I started. I have heard about bands wanting their CDs mastered as "hot" as possible, usually pop bands since they don't care as much about what happens to dynamics as much as they want their CD to be as loud or louder than other CDs when played on the radio. The fat cat's assumption has been that if your CD is louder on the radio, you will be more likely to sell more copies. But until you said this here, I was not aware of the presence bump. And that explains why other CDs, probably anything but pop, rock or c&w, where the musicians are more concerned with quality than with their product being "hot", can sound better (at least as good as is possible in 16/44.1). Can I assume that jazz and the less commercial material does no longer suffer from the presence bump or is this something that the labels still require? (And to think I spent 30 bucks on the book and the presence bump thing was not mentioned...at least not yet! I 'm about half way through it.) Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.