MP3 (320 kbps) VS flac/wav

[eggontoast]

I'd still have a tenner on Steve.

<pedantic spell mode>

And the word you're looking for is "lose"

</pedantic spell mode>

lol thanks for the correction, double posting, spelling error.....twice, must have been running in full plank mode.

A tenner doesn't scream like you have much faith, make it a ton and we can start to dance a little.

 

[Anonymous]

I'd still have a tenner on Steve.

<pedantic spell mode>

And the word you're looking for is "lose"

</pedantic spell mode>

lol thanks for the correction, double posting, spelling error.....twice, must have been running in full plank mode.

A tenner doesn't scream like you have much faith, make it a ton and we can start to dance a little.

Nah, I'm more into friendly bets. In fact, make it 1 dollar like in "Trading places"

 

[CnoEvil]

A couple of weeks ago, I went with a mate to listen to some "high end" equipment at his favourite specialist dealers, as they were holding an open-house demo day. There was some very expensive stuff being used there.

After listening to some very good recordings being played from vinyl and some very good sounding CD's, I was asked if I wanted to hear anything. As I'm curious about this Hi-res digital thing, I asked if they had any suitable music files so I could hear what all the fuss was about. I admit to having been rather sceptical.

A series of tracks were loaded into a playlist (I had a hand in choosing them) and, iPad in hand, I played a number of them through a Linn Akurate DS and an even more expensive non-Linn amp and Shahinian speakers.

Initially the first few 16/44.1 rips just sounded quite good or nothing particularly striking, but then I selected a couple of hi-res downloads versions of identical tracks that had been ripped to 16/44.1

Switching from a CD rip at 16/44.1 to the same track downloaded from a certain web site in 24/44.1 I expected to be listening out for some subtle difference. Heck no! It was immediately obvious that the higher res format was providing a much more open and detailed sound. I switched back and forth a few times to confirm what I was hearing and the difference was definately there.

The exercise was repeated for another track (acoustic - piano and female vocalist) with similar results.

A couple of vinyl diehards were listening too and they seemed to be quite astonished by this.

I tried a couple of 24/96 tracks and again was very impressed with the results, although I have to say my socks were almost blown off by an high quality recording on CD and by a particular "audiophile quality" vinyl LP.

The impression I got was that there's definately something worth persuing with the hi-res digital formats, but as always it depends on the source material.

We're repeatedly being told it's impossible to hear any difference, or that only some people will hear subtle differences some of the time. I had an open mind on it, but the difference I heard that day was quite obvious and I have no claim to "golden ears" or suchlike. Whether there's any advantage to hi-res files on mid-range or budget equipment might be another matter entirely?

This episode reminded me of certain experts in the 1990's telling us that all CD players sound the same....?????

.

Good to get your viewpoint. You have just mentioned the other (oft forgotten) variable in this debate, aside from imagination, the quality of the recording, intelectualization and deafness.....the quality of the equipment and how revealing it is. I suppose it's like saying that the bigger the screen, the more obvious the improvement of 1080p becomes.

 

[Anonymous]

Good to get your viewpoint. You have just mentioned the other (oft forgotten) variable in this debate, aside from imagination, the quality of the recording, intelectualization and deafness.....the quality of the equipment and how revealing it is. I suppose it's like saying that the bigger the screen, the more obvious the improvement of 1080p becomes.

I always think it's a very tenuous connection to compare video and audio like this.

CD is considered lossless (up to 20 Khz). The Nyquist theorum backs this up. All that is missing are those freqences (above 20 KHz), ie the sample rate is enough to "perfectly" reconstruct the original analogue waveform, all the way up to 22500 Khz.

FLAC is a lossless way to pack PCM. ie it's identical when played back.

MP3/AAC etc is a lossy way to pack PCM. Information is lost, but at high bitrates only information that is inaudible is lost (that's the theory).

Doing an ABX on a 320 MP3 v a FLAC, for me is impossible. The files sound identical (I appreciate some people reckon they can do it, but again, I've yet to see that...But I'll be very impressed if it's managed, and publicized.

However TV pictures are missing so much information that a 70 yr old with cataracts could tell apart a picture on a TV of an apple (say) and a real apple.

Yes, when we sit 2 or 3 metres away, it's good enough, but until we get pixel depths that are invisible to the naked eye, we haven't got anywhere near the equivalent quality of even a lowish bitrate MP3.

 

[CnoEvil]

I always think it's a very tenuous connection to compare video and audio like this.

CD is considered lossless (up to 20 Khz). The Nyquist theorum backs this up. All that is missing are those freqences (above 20 KHz), ie the sample rate is enough to "perfectly" reconstruct the original analogue waveform, all the way up to 22500 Khz.

FLAC is a lossless way to pack PCM. ie it's identical when played back.

MP3/AAC etc is a lossy way to pack PCM. Information is lost, but at high bitrates only information that is inaudible is lost (that's the theory).

Doing an ABX on a 320 MP3 v a FLAC, for me is impossible. The files sound identical (I appreciate some people reckon they can do it, but again, I've yet to see that...But I'll be very impressed if it's managed, and publicized.

However TV pictures are missing so much information that a 70 yr old with cataracts could tell apart a picture on a TV of an apple (say) and a real apple.

Yes, when we sit 2 or 3 metres away, it's good enough, but until we get pixel depths that are invisible to the naked eye, we haven't got anywhere near the equivalent quality of even a lowish bitrate MP3.

It was purely meant as an analogy, to illustrate that magnifying the difference (by highend gear), makes it easier to hear/see....nothing more.

 

[Anonymous]

Switching from a CD rip at 16/44.1 to the same track downloaded from a certain web site in 24/44.1 I expected to be listening out for some subtle difference. Heck no! It was immediately obvious that the higher res format was providing a much more open and detailed sound. I switched back and forth a few times to confirm what I was hearing and the difference was definately there.

The problem with these kind of comparisons (as mentioned in another thread?) is that you can't compare bitrate/depth this way. Most of the hi-res stuff has been mixed by an audio engineer with actual skills, so any difference you hear could very well be due to the mastering and not the format. If you really want to compare bitrates, you need to downsample the same file and compare that to the original.

CD is considered lossless (up to 20 Khz). The Nyquist theorum backs this up. All that is missing are those freqences (above 20 KHz), ie the sample rate is enough to "perfectly" reconstruct the original analogue waveform, all the way up to 22500 Khz.

Although the format is lossless, the quantization certainly isn't: the amplitude (volume) of the signal is reduced to a linear 16-bit value. You may have missed though, that the files were 24/44.1 so the Nyquist theorem still holds.

edit: not to detract from your original points: hi-res content usually is better since it needs to justify the price premium. I'm just not convinced it's really down to the bitrate. Likewise, I'd be very interested in a comparison 24/96 against 24/44.1, especially for people that already can distinguish between 24/96 and 16/44.1. For me personally, I consider LAME -v3 (variable bitrate, 192kbit) good enough for casual listening, but my own CDs are all stored in FLAC; storage space is cheap, Internet bandwidth not so much.

 

[Anonymous]

edit: not to detract from your original points: hi-res content usually is better since it needs to justify the price premium. I'm just not convinced it's really down to the bitrate. Likewise, I'd be very interested in a comparison 24/96 against 24/44.1, especially for people that already can distinguish between 24/96 and 16/44.1. For me personally, I consider LAME -v3 (variable bitrate, 192kbit) good enough for casual listening, but my own CDs are all stored in FLAC; storage space is cheap, Internet bandwidth not so much.

I agree. Having bought and downsampled some HD files myself, I personally think they are audibly indistinguishable. (I previously was convinced they were better, but I agree, it's just the mastering. Whether this is a deliberate "knobbling" of lower resolution files to make sure the high resolution versions are better, or simply expectation bias, I don't know, but it's definitely worth downsampling and seeing for onesself, like I've done. No more 24/96 and above for me.

 

[gowiththeflow]

The problem with these kind of comparisons (as mentioned in another thread?) is that you can't compare bitrate/depth this way. Most of the hi-res stuff has been mixed by an audio engineer with actual skills, so any difference you hear could very well be due to the mastering and not the format.

This was my thinking at the time; that the remaster was of a higher quality.

If that was the case with all hi-res download files, then all to the good IMHO. Anything that improves audio quality is worth it.

.

 

[John Duncan]

No more 24/96 and above for me.

Unless they were the same price as lower-res files and disk space was not an issue, presumably? I mean whether the differences are audible or not, I assume you'd have the master for reference and downsample for various uses?

 

[Andrew Everard]

I personally think they are audibly indistinguishable.

You mean you can clearly hear that you can't hear any difference?

or simply expectation bias

Try new Expectation Bias – when Placebo just isn't enough...

(cue sound of two arrows thudding into wood)

 

[Overdose]

With high resolution files, what is the most important component that gives the most noticeable improvement in sound?

Is it the bit depth, or the sample rate increase? As I understand it, the bit depth improves dynamics, but the sample rate would perhaps improve clarity. Is this way off the mark?

 

[Anonymous]

[EDITED BY MODS - please do not discuss moderation]

 

[MajorFubar]

Well I'm probably into the realms of conjecture again, but I'd say a faster sample-rate is more important than a greater bit-depth. In analogue-recording terms it's like increasing the speed of the tape (well sort of), where bit-depth is more like increasing the track-width (well sort of).

Why should it make a difference? Firstly, the Nyquist rule isn't wrong, it's just that - for example - a 10kHz tone sampled at 44.1kHz has only four distinct samples per cycle. With clever filtering that's enough to get a decent approximation of the sound on replay, but it's still pretty obvious that a greater number of samples in the first instance will do a better job. And real music isn't anywhere near as simple as a solitary 10kHz sine-wave.

That's my take on it, anyway.

 

[Anonymous]

No more 24/96 and above for me.

Unless they were the same price as lower-res files and disk space was not an issue, presumably? I mean whether the differences are audible or not, I assume you'd have the master for reference and downsample for various uses?

Absolutely. If 24/96 or higher became standard, and cost no more, I'd probably choose it, as you say, assuming disk space wasn't an issue (and that Wi-fi had come on a bit).

 

[Anonymous]

Well I'm probably into the realms of conjecture again, but I'd say a faster sample-rate is more important than a greater bit-depth. In analogue-recording terms it's like increasing the speed of the tape (well sort of), where bit-depth is more like increasing the track-width (well sort of).

Why should it make a difference? Firstly, the Nyquist rule isn't wrong, it's just that - for example - a 10kHz tone sampled at 44.1kHz has only four distinct samples per cycle. With clever filtering that's enough to get a decent approximation of the sound on replay, but it's still pretty obvious that a greater number of samples in the first instance will do a better job. And real music isn't anywhere near as simple as a solitary 10kHz sine-wave.

That's my take on it, anyway.

Major,

There are many fields in engineering where a layman's common sense will allow a good grasp of the fundamentals. Digital audio isn't one of those fields. Mental images of sinewaves turned into little staircases by sampling and the feeling that 'more samples are better' doesn't fully capture what is going on.

A 10KHz tone sampled at 44.1KHz will give a perfect rendition, not a decent approximation. The fact that it is 'pretty obvious a greater number of samples will do a better job' is layman's common sense conjecture getting in the way of the math.

As you rightly point out, music isn't a 10KHz sine wave, but what it is (for the purposes of digital audio), is a complex signal that is band limited to 20KHz. When sampled at a frequency greater than 40KHz, all the information will be captured.

So, you now have a mental image of a sine wave with a little squiggle on the peak of one cycle - how does that get captured? That little squiggle can be thought of as being made up of a combination of sine waves of different frequency - its frequency components. As the signal is band limited, that squiggle cannot have any frequency components above 20KHz - we know that because we have filtered them out. We also know that if we sample at least twice the maximum frequency we can recreate all the component frequencies of the squiggle, and therefore the squiggle gets fully captured.

I think the problem comes from not mentally putting our imaginary spiky 'sampling won't get this signal right' waveform through an imaginary low pass filter first before we do our imaginary sampling.

Higher sampling frequency does have benefits - most notably it gives an easier time to the analogue filter stages after the DAC, but that is nothing to do with more samples doing a better job, just the realities of analogue filter design.

 

[Anonymous]

With high resolution files, what is the most important component that gives the most noticeable improvement in sound?

Is it the bit depth, or the sample rate increase? As I understand it, the bit depth improves dynamics, but the sample rate would perhaps improve clarity. Is this way off the mark?

Increasing bit depth will improve the dynamic range of the system, and increasing sample frequency will increase the maximum frequency the system can reproduce (if that is what you mean by clarity).

Worth pointing out that Philips didn't hit upon the redbook standard by accident, good old 16/44.1 CDs have a higher frequency response and wider dynamic range than human audio perception.

It is open to question whether higher sample rate / greater bit depth does indeed improve the sound over a well recorded and implemented redbook standard CD. Many comparisons have used different starting points - a 24/96 download may well have come from a different master than the 16/44.1 CD. The safest way to do it is to have a 24/96 download and then downsample yourself to 16/44.1 and compare the sound of the two files. I have tried this through a decent system and was unable to tell the difference - but I am an old sceptic who cant hear above 14KHz anymore.

 

[steveozs]

I personally think that if anyone is going through all the trouble of getting your system sounding and being as good as it can be why on earth would you make the comprimise to save on a few mb's per file?

I completley agree with this statement. If I wanted to rip a CD collection from scratch then the obvious choice is to use lossless because hard disk space is so cheap. If you also want some MP3's to use with a portable player then you can easily batch convert a second copy of the music.

Some 320kbps files do sound very similar if not pretty much the same to some lossless files, some sound noticeably worse, it all depends on the recording to start with, but with classical and especially Jazz there is more depth to the sound (in my experience) and the only way I can describe it is more of a '3d' type effect to the music. Either way my preference is as I said to go with lossless, there's no point in crying over a bit of storage.

I used to think the same thing as you until I tried an ABX test. When you compared the MP3's to lossless was it just a 'sighted AB' test or have you ever tried doing a proper scientific 'ABX' test? Download Foobar2000 and install the ABX plugin and I bet you won't be able to tell the difference between 320kbps MP3 and lossless files (assuming both files are ripped from the same a CD of course).

I'll give that a try! Cheers.

 

[steve_1979]

I've just tried some 'lossless vs MP3' experiments on a friend of mine and got some very interesting results.

For the first experiment I played him a series of two identical MP3 files but lied and said that one of them was lossless. Most of the time he couldn't tell any difference but a few times his said the at the fake lossless file was better than the MP3 even though it was really identical.

For the second experiment I swapped the files around. I told him that the MP3's were lossless and that the lossless files were MP3's. This time he thought that the MP3's sounded better than the lossless files.

For the third experiment I played him some lossless and MP3's and told him the truth about which one was which. He mostly prefered the lossless files to the MP3's this time.

For the fourth experiment I let him do an ABX test using Foobar and he couldn't tell the difference.

Has anyone else ever tried any tests like this? I think the results that I got are pretty suggestive and I'd be very interested to find out what results other people get.

 

[steve_1979]

I'll give that a try! Cheers.

I look forward to hearing about how you ABX tests go.

 

[MajorFubar]

Major,

There are many fields in engineering where a layman's common sense will allow a good grasp of the fundamentals. Digital audio isn't one of those fields. Mental images of sinewaves turned into little staircases by sampling and the feeling that 'more samples are better' doesn't fully capture what is going on.

A 10KHz tone sampled at 44.1KHz will give a perfect rendition, not a decent approximation. The fact that it is 'pretty obvious a greater number of samples will do a better job' is layman's common sense conjecture getting in the way of the math.

As you rightly point out, music isn't a 10KHz sine wave, but what it is (for the purposes of digital audio), is a complex signal that is band limited to 20KHz. When sampled at a frequency greater than 40KHz, all the information will be captured.

So, you now have a mental image of a sine wave with a little squiggle on the peak of one cycle - how does that get captured? That little squiggle can be thought of as being made up of a combination of sine waves of different frequency - its frequency components. As the signal is band limited, that squiggle cannot have any frequency components above 20KHz - we know that because we have filtered them out. We also know that if we sample at least twice the maximum frequency we can recreate all the component frequencies of the squiggle, and therefore the squiggle gets fully captured.

I think the problem comes from not mentally putting our imaginary spiky 'sampling won't get this signal right' waveform through an imaginary low pass filter first before we do our imaginary sampling.

Higher sampling frequency does have benefits - most notably it gives an easier time to the analogue filter stages after the DAC, but that is nothing to do with more samples doing a better job, just the realities of analogue filter design.

I'll ignore the fact that your post came across as slightly condescending (perhaps it didn't intend to), and I will just say: thanks for your in-depth explanation, but I will stick to my belief that there are obvious theoretical benefits to be had from increasing the sampling-rate beyond 44.1, particularly with high-frequency sounds. Those benefits largely revolve around the reasons that I described, even if I used 'laymans common sense' to describe them.

There isn't a recording-studio out there which still records natively in 16/44: it's considered too low-def, even if the final master of an album is downsampled to it for CD purposes. I'll tell the several recording-engineers I am acquainted with that other than allowing the audio-filters to chill-out a bit more, by and large they're wasting their time recording their clients at anything beyond 16/44.

 

[Anonymous]

There isn't a recording-studio out there which still records natively in 16/44: it's considered too low-def, even if the final master of an album is downsampled to it for CD purposes. I'll tell the several recording-engineers I am acquainted with that other than allowing the audio-filters to chill-out a bit more, by and large they're wasting their time recording their clients at anything beyond 16/44.

There are very good reasons to "record" in higher resolution, and that's because of multiple mixes, edits, masters, chopping music around. All this can introduce errors. The higher the resolution, the smaller the errors will be.

The final CD at 16/44.1 gives an "exact" representation of the analogue so long as no frequencies above 20 Khz or so are present. He explained it pretty well.

Imagine it like editing a photograph. A high quality JPEG is undistinguishable from a bitmap on a computer screen, but ideally you "edit" the RAW file and save as JPEG. If you want to edit again, you go back to the RAW (or usually PSD (photoshop)), because all changes to the JPEG are lossy. So the more you edit, and re-edit the more potential errors creep in.

So when you want to make a new mix, master of a recording, you go back to the higher resolution version, do your stuff, and record to CD at 16/44.1

 

[John Duncan]

How is classical music (one of the big 24/96 genres) recorded? I always imagined it being pretty much live in one take?

 

[MajorFubar]

The final CD at 16/44.1 gives an "exact" representation of the analogue so long as no frequencies above 20 Khz or so are present.

With all due respect, I suspect that one sentence there is one area where we are not destined to agree, so I will not labour my point any further, other than to say, I personally do not believe that to be the case.

How is classical music (one of the big 24/96 genres) recorded? I always imagined it being pretty much live in one take?

It depends: often there can still be overdubs and splices from different takes to create a final mix. If someone messes-up near the end of a long piece, it's cheaper to just re-record the offending bit than make them re-record the whole performance again. Orchestras tend to be too expensive to make them play everything again unnecessarily.

 

[Anonymous]

The final CD at 16/44.1 gives an "exact" representation of the analogue so long as no frequencies above 20 Khz or so are present.

With all due respect, I suspect that one sentence there is one area where we are not destined to agree, so I will not labour my point any further, other than to say, I personally do not believe that to be the case.

Major, You are of course free to personally believe what you like. The point I was trying to make earlier was that a layman's intuition can in many cases lead to the wrong conclusion.

There are entire industries built upon the bedrock of the Nyquist-Shannon sampling theorem, the phone in your pocket, the music on your ipod, the electronics in your car. I think it is likely that if a flaw existed, it would have been spotted by now.

Have a search on the internet and see if you can find theoretical support for your viewpoint.

On the other hand, you could try reading the following, written by Dan Lavry, which tries to address the points you raise without delving into too much math. I would recommend it.

http://www.lavryengineering.com/documents/Sampling_Theory.pdf

 

[Anonymous]

Good article Andy. Explains it well...

@Fubar.

The only plausible reason I've seen to suppose higher samping rate for playback is better quality than 44.1 KHz is the argument that you get all the extra information above 20 KHz, whch although inaudible, can possibly be felt, or that their interaction with lower frequencies can have an affect on harmonics lower down.

Extra bits is irrelevant. The 16 bits gives us a dynamic range of 96 dB, which is far higher than any recording, and in fact if utilised, then played back so we could hear both the quietest AND loudest parts of the music above the noise floor would cause permanent injury to the ears!