"Everything between sample points is lost"
Mmmm. Messrs Nyquist and Shannon might have a bit to say about this.
How the FLAC do I tell MP3s from lossless audio?
Can people hear the difference between lossy MP3 digital music files and lossless ones? Opinions differ strongly, with much obfuscation around audio cables, mastering and hi-fi componentry muddying the waters. This article was prompted by Reg reader critiques of Sonos streaming Wi-Fi speaker/player reviews and audiophiles …
-
-
Friday 26th September 2014 18:02 GMT the spectacularly refined chap
Re: "Everything between sample points is lost"
Mmmm. Messrs Nyquist and Shannon might have a bit to say about this.
What, you mean to agree with him? There is no error there: it stands to reason that if you are ignoring the input at any given point what happens during that time cannot be passed through to the output.
-
Friday 26th September 2014 18:09 GMT ThomH
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
I think he means to make the distinction between the time domain and the frequency domain. Assuming perfect instantaneous sampling then everything in between samples in the time domain is lost. But, frequency wise, there's no new information between samples to miss.
I guess it depends on what you define the totality of the information to be. If you want to record all frequencies up to 20Khz then if you have regular samples at 40Khz, nothing is lost. If your low-pass filter is insufficient then aliasing may even add things that weren't there originally...
-
Friday 26th September 2014 18:24 GMT the spectacularly refined chap
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
I think he means to make the distinction between the time domain and the frequency domain. Assuming perfect instantaneous sampling then everything in between samples in the time domain is lost. But, frequency wise, there's no new information between samples to miss.
But that's the whole point - you have defined a frequency domain. A real life audio signal does not keep to neat boundaries so something like a clash of cymbals for instance will reach well into ultrasound territory. If you are sampling at 44.1kHz that is going to be lost. The fact you are defining a region of interest - presumably some "human hearing" range - is itself an acknowledgement of that. The data is lost regardless of whether you were interested in it or not.
-
Friday 26th September 2014 19:00 GMT Paul Crawford
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
The key point about Nyquist's theorem is it starts with the assumption that the signal you are interested in is strictly limited in bandwidth. If that initial assumption is true, for example that you only want/need 20Hz to 20kHz, then by sampling above twice the highest frequency (say at 40.0001kHz) than you are NOT losing any information by sampling.
What is impotent is that 20kHz is an arbitrary value (but realistic limit for most younger humans, us old buggers are lucky to get 15kHz) and to avoid the very unpleasant business of aliasing you MUST be strictly limited to that value.
Since that near brick-wall filter is highly impractical for any analogue filter, what is normally done is to sample higher than that, either a little bit more on sample rate (like 44.1kHz) and use good analogue filters, or a much, much higher sample rate and push the band-limiting problem in to the digital domain where it is practical to implement good filters (but with time delay, but for recording that in not a problem) and then to re-sample at a chosen lower rate.
-
This post has been deleted by its author
-
-
Friday 26th September 2014 19:24 GMT Unobtanium
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
"ultra"-sound meaning beyond sound, I presume?
You can't hear it; that's why we only pass up to about 20 kHz into the recording chain. If it's not present in the input, its not "lost" from the output. Nyquist's theorem is the law. The reproduction of a low-pass signal is essentially perfect - nothing is "lost".
Of course, no real-world implementation is perfect, but there's no big mathematical or philosophical discussion hiding here.
-
-
Friday 26th September 2014 19:24 GMT Anonymous Coward
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
that's a bit of a muddle - if the signal has no energy above some frequency and you sample at least twice that fast - and the output filter is good enough - the output waveform is (essentially) identical to the input. Nothing is "lost".
-
This post has been deleted by its author
-
Saturday 27th September 2014 16:09 GMT Unobtanium
Re: "Everything between sample points is lost" (@the spectacularly refined chap)
"FOR FUCK'S SAKE HAS NOBODY EVEN BOTHERED TO READ THE ARTICLE?
Read the quote again: "Everything between sample points is lost". If there is a signal faster than that it is gone. That is not conditional on anything. You can argue about whether it is relevant or not but it is gone never to be recovered. That is precisely what Shannon said."
Strident, profane, but wrong. If you understood sampling theory you'd know that frequencies above half the sampling rate are not "gone"; their energy is aliased back down into the baseband (the spectrum is effectively "folded" back on itself). It's actually more complicated than that, but that's close enough for this discussion. The article's phrasing is unfortunate, at minimum, but I lean toward "wrong".
-
-
-
-
Friday 26th September 2014 22:36 GMT Steve Knox
Re: "Everything between sample points is lost"
There is no error there: it stands to reason that if you are ignoring the input at any given point what happens during that time cannot be passed through to the output.
This is true if and only if there is no deterministic relationship between the samples and the unsampled data (i.e, there exists no function f where f(s) = u.)
-
Saturday 27th September 2014 14:06 GMT Spanky_McPherson
Re: "Everything between sample points is lost"
This sentence is fundamentally wrong. Nothing between the sample points is lost. The continuous analogue signal can be perfectly recreated from the samples.
Please please watch this excellent video made by someone more knowledgable than anyone on this forum: (watch between 4:00 and 6:00 if you don't have time to watch the whole thing)
https://www.youtube.com/watch?v=cIQ9IXSUzuM
(For the pedants, yes, this assumes that the signal being sampled does not contain frequencies above 22.05kHz. Obviously this filtering is always done to the signal prior to sampling)
-
-
Friday 26th September 2014 18:46 GMT Rebecca M
Re: "Everything between sample points is lost"
Mmmm. Messrs Nyquist and Shannon might have a bit to say about this.
I'm sure you feel such a big boy quoting those names. Pity that it doesn't automatically make you right or knowledgeable, indeed it simply shows that you missed their central tenet. Encode a 100kHz signal at 44.1kHz and then regnerate the wave from the sampled data. That 100kHz signal is not present in the output. If it hasn't been lost then where has it gone. That is the whole point of Shannon-Nyquist - the sampling frequency determines the maximum frequency that can be sampled.
From there you get to events that occur faster than the sampling frequency can't be captured which remains true however those samples are captured - I see another poster is bringing in whether the samples are instaneous readouts or integrations which is an utter irrelevance - the principle holds regardless of the sampling methodology.
The article states that events that happen faster than the sampling frequency can't be represented. That is true. So again, precisely what is wrong with that quoted text?
-
This post has been deleted by its author
-
-
Saturday 27th September 2014 07:16 GMT TheOtherHobbes
Re: "Everything between sample points is lost"
Nope. Sampling includes filtering to get rid of the aliased copies. If it didn't, it would sound really, really horrible.
As other posters have pointed out, the filtering is never perfect, so sampling is never Nyquist-perfect either. Sampling at higher frequencies and higher bit depths should have fewer imperfections, although if the hardware is a bit pants anyway high-rate sampling won't make a huge difference. (And if it's really pants it can make the sound worse).
But the killer problem for digital is clock jitter. If the sampling clock isn't rock solid to nano-second precision, you can forget Nyquist, because Nyquist assumes perfect sample timing. A lot of the smeary-splashy-nasty sound digital used to be famous for was caused by cheap jittery clock sources.
If your DAC can accept an external clock, hooking up a studio-grade clock source will do the sound many favours. It will also make the differences between FLAC and MP3 more obvious.
IME I can hear the difference very clearly, and the MP3 sound is seriously fucking annoying, even in a car. But my gf, who is a classical musician and can pick out the notes in chords by ear, is fine with MP3s. She hears music as pitch lines and very fine timing details, and most timbres as a placeholder. MP3s include all the detail she needs.
In fact everyone hears differently anyway, because everyone's ears are a slightly different shape, so we all have different acoustic filters stuck to sides of our heads. So it's maybe not a surprise some people strongly prefer FLAC and others don't care.
Last point - CDs aren't really lossless. Because of dirt, scratches, laser servo issues and other inaccuracies, most players drop back to Level 1 error correction at least some of the time, so there's always some quality loss.
A good CD rip will be bit-perfect with multiple read passes made to minimise errors, so FLAC will always sound better. (I was amazed by the difference - so amazed I spent a few months ripping and selling off all my CDs.)
-
Monday 29th September 2014 08:23 GMT mr_z
The frequency domain adds pre-echo
If you just sample a low-pass filtered signal, even with a bit of jitter, and play it back, you won't add appreciable artifacts. Yes, there will be some, but the artifacts inherent in MP3 and related algorithms are orders of magnitude higher.
MP3 divides the signal into frames, and performs a Modified Discrete Cosine Transform to the signal. This transforms the signal from the time domain to the frequency domain. Then, it compresses the MDCT coefficients by quantizing them, guided by a psycho-acoustic model.
(Psycho-acoustic model means: "We've algorithmically determined that you can't here this thing we're throwing away." It's based on many studies of the masking effects inherent in human hearing, such as not being able to hear certain sounds after a loud plosive sound, etc.)
Quantizing in the frequency domain adds non-causal artifacts to the signal. What do I mean by "non-causal"? You can get what some call a _pre-echo_ before sharp time-domain discontinuities in the input, such as percussive sounds. Pre-echo is what makes percussion sound "muddy" or "blurred". You start to hear a snare hit or cymbal before it's been hit.
That's why I call it non-causal: Analog filtering and properly designed digital filtering don't change the leading edge of a discontinuity; rather, there's an impulse response that appears after the discontinuity. But, with frequency domain quantization, the artifacts get spread to both sides.
You've likely already experienced this elsewhere: highly compressed JPEGs and MPEG video! Take a look at what JPEG and MPEG do to areas of sharp contrast, such as text. You see "sparkles", "ringing" or "mosquito noise" to all sides. Both are based around a similar frequency domain transform, the DCT, and both perform similar quantization, only in two dimensions (horizontal and vertical) rather than one (time).
But the artifacts arise from the same place, mathematically.
If you read the design documents on Ogg Vorbis, they're very sensitive to the issue of pre-echo.
There are other artifacts I can hear in MP3 (especially heavily-compressed MP3) that others don't notice. There's burbles, the occasional tone that sounds like Morse code, and so on. These too are artifacts of popping to the frequency domain and quantizing frequencies to varying degrees.
As for the idea that "most people hear differently:" Because I've worked with our digital video folks, I'm quite sensitive to video artifacts, including DCT artifacts, but also spatial domain quantization (resulting in "contouring") and so forth. My wife and friends never really noticed many of these until I started pointing them out. Now they hate me for "ruining" them. ;-)
All that said: I can definitely hear artifacts in 128kbps CBR recordings, fewer in 192kbps VBR recordings, and rarely or never in 256kbps or 320kbps. At 320kbps, you're only compressing about 3.5:1 or so, and so you're leaving most of the signal intact.
Likewise, I rarely notice JPEG artifacts on something compressed with 90% or higher quality, but then the compression rate also drops significantly compared to lower quality levels. At that point, if it has a lot of text, you may be better off with PNG anyway.
-
-
-
-
This post has been deleted by its author
-
-
Friday 26th September 2014 19:00 GMT 142
> But if you can't tell the difference already, what "breakthrough" could possibly improve the experience?
Well there a lot of things you might want to do spacially, or spectrally, with the sound to spread it across different speakers, etc.
The problem is, once you encode audio with a perceptual codec like MP3, the resulting audio makes sense to the human ear in a very specific set of circumstances: that it's played in stereo through a standard listening system.
Outside of this, it breaks down.
The reason is straightforward: MP3 primarily removes parts of Sound A because Sound B has played before, or simultaneously, making some parts of Sound A inaudible ("masked").
Now, if you go changing the sound, dividing it between speakers, extracting elements, etc, the Sound B is played back differently and hits the ear differently, meaning different parts of Sound A get masked, and the parts the MP3 cut out of Sound A become obvious (it sounds akin to an MP3 with a 64kbps bitrate, or crap YouTube vid, with all the swirling noise and distortion).
Similarly, if you need to digitally analyse the sound for some reason, the loss of data is very significant, as it was encoded to sound good for a person, but the signal doesn't make sense in its own right - the removal of all those chunks of audio looks like weird noise added to the signal.
-
Friday 26th September 2014 19:02 GMT Paul Crawford
The ability to tell the difference depends on 3 things:
1) The original quality of the recording.
2) how good your system and ears are.
3) What sort of MP3 compression is in use.
Number (3) is critical, if you are using 128kbit fixed-rate coding then I am pretty confident you will tell the difference, if you are using 320kbit variable-rate I would doubt most could.
-
Saturday 27th September 2014 09:50 GMT Mage
ability to tell the difference depends on 3 things
Also probably 6" drivers in MDF, chipboard, or possibly plywood baffles / boxes.
Not 4" drivers in a plastic case, which can often be poorer than £2 ear buds. Over the Ear 'phones' actually are harder to do properly than "in ear" earphones.
-
Saturday 27th September 2014 10:46 GMT Dave Bell
Re: ability to tell the difference depends on 3 things
My experience is with "home cinema" situations. Most of the cheap rigs you can buy for that have small speakers for the directional channels—the surround sound—and a subwoofer. But you can hear when all the bass is coming from the front, rather than from the same direction as the high-frequency element.
My father's hearing was pretty bad, but having background noises coming from different directions than the character speech made a big difference. Even just putting the normal TV sound through a Dolby Analogue Surround decoder can help (It makes me wonder what some of the "bad" TV sound really is.)
Anyway, most stereo recordings position sound sources merely by relative volume. They don't bother with phase. It's a trick. You don't hear anything now of binaural stereo, which did try to capture the phase differences. The modern digital surround sound does stand a chance of using that data, but speaker design tends to ignore frequency-related phase shifts.
Your show-off hi-fi audiophile with the big speakers isn't exactly wasting money, not until he gets into exotic speaker cables, but he may be missing out on a lot by sticking with such an ancient sound recording standard. Stereo sound is still faking it. So might a 5:1 recording, but it doesn't have to throw away so much data
-
-
-
Friday 26th September 2014 19:35 GMT Anonymous Coward
CDDA < FLAC
In short, if your source is 96khz or higher, try something like FLAC. If it is CDDA, your wasting space with FLAC. That opinion is not from someone who is a technical expert on sound, but from someone who has remastered 100's of songs.
FLAC is severe overkill when your input is CDDA. There is a much more significant visual difference between BetaMax and VHS than the audible difference between FLAC and MP3@256 if your source is CDDA (mp3@320 and you're sort of wasting space again). So, how much visual difference is between Beta and VHS again?
In 2003 I started remastering individual songs that had never been remastered using a plethora of DSP software (no big boy hardware :-( ). If I couldn't find a studio master (which were far easier to find then), I would choose CDDA. At some point, I came across a master of Moon Safari by Air (not for remastering, but because I liked it). The master of course didn't sound the same as the CD I had, so I looked for my CD for comparisons, but I couldn't find it. In desperation to find the differences by comparing the master to my CD, I substituted .mp3 files for my CD as the source.
A week passed and the differences became apparent, however I still wasn't using CDDA, so I acquired the CD version to ultimately compare CDDA to .mp3. What I found between my 256.mp3's and CDDA was eye opening. Not only was 99.5% of the audible information there, but more importantly, the information that was NOT in the CDDA but in the master, was also not present identically in the 256.mp3's, and the .mp3's were sourced from CDDA! If CDDA didn't have it, .mp3 didn't have it identically.
You really can't remaster from CDDA, but whatever you can make better out of a CDDA, you could do just the same from decompressing a 256.mp3. I lived it.
-
-
Friday 26th September 2014 21:10 GMT Anonymous Coward
Re: CDDA < FLAC
Supposedly, Nilesh Patel. However, I'm positive that he wasn't the one that put it "out there" back then. It was probably some intern in broadcasting at the label or something similar to that in this case, or in all cases truthfully. I haven't looked for any masters in years, but last I looked about 6 years ago, I couldn't find 1 person that had any, new or old.
BTW, here's this if you care http://www.dummymag.com/news/legendary-mastering-engineer-nilesh-patel-dies
-
Saturday 27th September 2014 03:43 GMT 142
Re: CDDA < FLAC
:-( Yeah, legend indeed. I'd have loved to have had the chance to send a mix his way.
Yeah, the reason I ask is that it seems bizarre to me that there would be any audible difference between the masters the mastering engineer made, and what was on CD, given that that would be the delivery format from the mastering engineer to the label in most cases. What leaves the mastering studio ends up on the CD with bit-perfect accuracy.
So if you were noticing a difference, I wonder were you getting the *mix masters*, from the mix engineer, before it got sent to mastering?
-
Saturday 27th September 2014 19:22 GMT Anonymous Coward
Re: CDDA < FLAC
@142 "...I wonder were you getting the *mix masters*"
Actually, those are what I wanted, but what I received was 1 file of a 2-channel mix almost 4GB in size. Sometimes these masters came in multiple channels, most times not (30/70). Being 4GB in size, I knew immediately it wasn't what I wanted, but how could I refuse :-).
At that time what I wanted was a proper 5.1 mix, I still would. It's been too long to remember the exact numbers of anything. The only things I remember are that it was 2 channels, it had a few tracks that were not on any release (at that time) and it had a much higher sampling rate. I think I read some years back that a new vinyl cut had been released, apparently the finest release of the album, but I never investigated.
I still wish that the RIAA would of made an initiative to forcefully archive higher quality versions of ALL albums that were digitally distributable in at least DVD-A spec (5.1 24b/96khz). DVD-A spec has been around nearly 20 years and is now considered archaic. However, it's still far better than CDDA in all aspects (doubley better). It makes me wonder why people consider loseless CDDA at all when CDDA itself is extremely lossy compared to so many alternatives (new and archaic).
For me, comparing CDDA to MP3 is exactly the same as choosing which flash drive to store text documents on: 16GB or 64GB? I'll take the one that gives more space.
-
-
-
-
-
Saturday 27th September 2014 10:01 GMT Dave Bell
MP3 is a pretty smart piece of sound engineering, dropping sounds that most of us can't perceive, but it would be foolish to think there can't be better.
If I had something on a lossless file on my desktop machine, and something came along that would measure my hearing and compress in a way that took my personal limits into account, the result might be better for my ears, and compress better than MP3.
Fanciful? Well, the big problem is the measurement of the response of my ears, but they only get up to about 12kHz these days, and compression suited to J. Random Teenager could include a lot I cannot hear, so maybe a personal compression standard would work.
But if there's something like that, we're going to need a lossless source for the compression.
-
Saturday 27th September 2014 22:54 GMT harmjschoonhoven
@Dave Bell, Re: MP3 compression
You might have a look at
http://www.harmjschoonhoven.com/mp3-quality.html#listen
where you can hear what MP3 (the MPEG1 Layer III audio codec, as defined in ISO 11172-3) discards by opening one of the links in the table with dBA values. It is pretty horrible.
My impression is that MP3 leaves the sound intact, but compromises the emotional impact. May be a boffin with access to an fMRI scanner can do some research on this.
-
-
Tuesday 30th September 2014 12:57 GMT Michael Wojcik
But if you can't tell the difference already, what "breakthrough" could possibly improve the experience?
And what if you don't care? I've been buying music (defined broadly) for three decades, and I consistently find myself unable to give a damn about fidelity. There are songs I enjoy listening to, and I enjoy them just as much from a bargain-basement MP3 player and earbuds as I do from a CD and fancy audio components (when I hear them played on someone else's system, since I don't own any player that cost more than $30).
Yes, I understand that many people do care; but some of the codec warriors don't seem to understand that not everyone shares their passion.
That said, when our phones all have terabytes of storage, we'll probably all use lossless.
I'll use whatever format it comes in when I buy it.
-
-
Monday 29th September 2014 03:10 GMT Mark 65
The reason I use FLAC? It's my archive copy. CDs get scratched, lost and broken. They also degrade over time, especially in hot climates. My FLAC copies are backed up in multiple locations. Versus the original CDs: they are searchable; they are streamable; they are shareable en masse. I can also re-encode them quickly to AAC for use on portable devices and when AAC is surpassed by some other format in order to cram yet more music onto small devices I will re-encode to that.
-
-
Friday 26th September 2014 17:30 GMT Anonymous Coward
"I'd love to get more listening pleasure from the music and hear it as the artist intended."
Go to live performances in the flesh. Unless they are miming then you get a different performance every time. You also get the rapport building between the performers and an engaged audience - and that's something you can't put in a bottle.
Afterwards when you play a recording your brain will override what you are actually hearing - by "replaying" the emotion of the concert. IIRC an academic study was done on this effect a while back.
-
-
Saturday 27th September 2014 03:26 GMT Tom 35
Re: goto a live ACOUSTIC performance
If that was true after seeing The Who yeas ago at the Toronto Sky Dome* I would hear the sound of an AM radio in the trash bin of a large tiled washroom now.
*lucky I won tickets from a local radio station contest and didn't pay the silly price that floor tickets were going for.
-
-
Saturday 27th September 2014 02:49 GMT JeffyPoooh
I did. It was traumatizing.
"Go to live performances in the flesh."
Apparently one should close one's eyes to listen to a orchestra. It prevents one from witnessing the brass section turning their horns over and over to drain copious volumes of spittle onto the floor, leaving glistening patches of bubbly drool there, shimmering under the lights. It was difficult to keep the bile in place. Music? What music? I don't recall any music. Just saliva. Night of the Saliva.
At least nobody bit into a freshly killed chicken.
Recorded music has its advantages. Fewer diseases.
-
-
Friday 26th September 2014 17:32 GMT Pete 2
Listening too hard
> I think I can detect an instantly perceptible MP3-FLAC difference
Maybe you can - but does it really matter?
Most people I know listen to music as a form of entertainment, generally as relaxation. They don't listen to it on the assumption that they will be tested on it's content and clarity after hearing it. Likewise, they don't listen, eagle-eared. waiting for that instance in the third passage where the conductor's tummy rumbles - or where you can hear the tube train rolling past the recording studio.
Having said that, the first time I plugged in my home-made transmission line speakers (still with me 30+ years later) and cranked up Wish You Were Here it was a bloody revelation. I have witnessed similar reactions when I have plugged in a basic 2+1 speaker system into friends' flat-panel tellies: where did all that sound come from? after listening to tinny audio for an age and not realising there was anything better.
Although those step-changes are huge. Whereas the difference between an average quality MP3 and a FLAC is perceptable - but you're merely detecting the difference, not listening to it. And as soon as someone in the upstairs flat farts, or a car rumbles past, the difference vanishes. As it also does on anything less than my TLs.
Biting the hand that feeds IT
