MP3 and audio data compression

[STHLS5]

Thanks Kittykat. I just realized my CD player may not be able to play FLAC or AAC.

I just finished burning Bach: BMV. 565 Toccata (for the deep organ sound) and Romance de Amour (for guitar) using 256kb MP3 and copied the same in wav format. I used Window Media Player to do this. Without knowing how the sequence of the tracks format, I managed to pick wav files to be better. However, whatever difference I can tell is only when it is played side by side for comparison. If I were to walk into the room and asked to identify which format is playing, I would not be able to tell.

Interestingly, when I listened to the same tracks using my computer, they all sounded the same to me. Maybe, I should have burned them at 320kb. I was told to use VBR setting, but I can't find that option in WMP.

ST

[weaver]

As some readers here will already know, the BBC is experimenting with a high quality internet stream for the remainder of the Proms concerts. This applies to the live broadcasts and the daily repeat programmes but not the on demand 'Listen Again' feature.

[It can be accessed here]

This stream is at 320 kbps as opposed to 192 kpbs for Radio 3 at other times.

There is also a blog attached to the project here.

Reading the comments that have been added it quickly becomes apparent that listeners are delighted with the improvement in quality but also that they believe the dynamic range of the broadcasts to be greatly improved.

It has been a criticism of Proms broadcasts, be that via TV, FM radio or internet, that the dynamic range of the concerts has been heavily compressed; this current stream appears to show a vast improvement in this area.

As I understand it, compressing the data prior to broadcast would not of itself lead to dynamic compression, but clearly the two are lumped together (in the listeners mind) in this instance.

I have no knowledge of the dynamic limitations of the various broadcast channels but my question would be: in the case of broadcast music (TV, FM radio, web) does dynamic compression necessarily go hand in hand with data compression?

[A.S.]

...I just finished burning Bach: BMV. 565 Toccata (for the deep organ sound) and Romance de Amour (for guitar) using 256kb MP3 and copied the same in wav format.

We're curious as to why you picked organ and guitar as a WAV v. MP3 comparison. As I've reported here before, the organ sounds good on every loudspeaker (it has a narrow frequency range and limited harmonics) and obviously the (acoustic) guitar has a wider frequency range but almost no dynamic range at all.

As Mod1 said, what are you actually trying to listen for? You know what the secret of a really good research test is? It's thinking through the likely outcomes, and constructing the test to maximise the comparative differences long before lifting a finger to actually perform the test. In this example, musical selection is of paramount importance.

[Labarum]

As I've reported here before, the organ sounds good on every loudspeaker (it has a narrow frequency range and limited harmonics) and obviously the (acoustic) guitar has a wider frequency range but almost no dynamic range at all.

Alan

Do all organ stops have limited harmonics? I am guessing reed stops are rather rich in harmonics and can have a rather steep starting transient.

Doe the guitar realy have greater frequency range than an organ? Can it reach down to CCCC, or sound as high as a 2ft Mixture Stop?

Of course the organ will test a speaker's bass like no other instrument, and some of the stops produce very complex sounds.

[A.S.]

Before I even bother to reach for the text book, let's apply some common sense. If you have a bi-wired pair of speakers, unhook the biwire link feeding the tweeter and play organ and then guitar. You'll notice that a) it hardly matters whether the tweeter is connected or not when playing organ b) if definitely very much matters if the tweeter is connected playing acoustic guitar. That's because the guitar has a complex harmonic structure which extends far beyond the plucked note.

That proves to my satisfaction that the guitar has a wider range in the region that we are most interested in for comparative testing i.e. 3-6kHz. It's not relevant what happens down at low frequencies for this sort of comparative test because the problem with all digital (linear or compressed) audio coding systems is never how it handles the bass (where the digital resolution is highest and the ear's sensitivity to distortion the lowest) but how it handles the middle and high frequencies where the digital resolution is lowest and the ears sensitivity the highest.

In other words, you could probably get away with 6-8 bit resolution in the bass, but you'll need 16 bits where the ear is most sensitive (3kHz). So MP3 will easily be good enough to handle the coding of an organ. It would be a trivial matter to make an MP3 recording of an organ sound acceptable even at (I dare say) not 320kb but 96kb, or even less to code the bass adequately*. Probably less because the sound is virtually monophonic, so could probably be dropped to 40-50kb. Try it yourself!

*It may be necessary to increase upwards from 96kb depending upon whether the organ music is concentrated in the bass or has more fiddly higher-register content.

[Pluto]

As I understand it, compressing the data prior to broadcast would not of itself lead to dynamic compression, but clearly the two are lumped together (in the listeners mind)

The two are frequently confused and interchanged, incorrectly.

For this reason I prefer the terminology

• Compression meaning dynamic compression as deliberately caused by a compressor, limiter, loudness enhancer, automatic gain control etc. or as a concomitant effect of analogue tape overload.

• Data reduction covering the effect of MPx, AAC etc.

[STHLS5]

We're curious as to why you picked organ and guitar as a WAV v. MP3 comparison. ..

I was hoping for some assistance from people who were familiar with MP3 for some suggestion but except for Kittykat I received no help. Being a non technical guy I relied on www.independentrecording.net to choose the most suitable recording to do the test. They didn't say to choose but at the frequencies chart in their website organ seemed to be the most appropriate and I love the bass.

1)According the chart organ fundamental frequencies extend from 10Hz to about 7.5kHz.

2) I chose guitar because that's one of instrument I like. That matters to me how MP3 is going to perform for my type of musical preference.

My so called "test" is the real world test for people like myself who judge how a format performs to our liking. Despite, whatever conclusion the audio engineers have about the transparency of 256kb MP3, I hear obvious difference in Bach BMV 565 Toccata compared to Romance de Amour.

If I am wrong, then the [COLOR="rgb(139, 0, 0)"]findings as published in AES[/COLOR] were also wrong and BS. Reproduced the abstracts for further reading.

Subjective evaluation of MP3 compression for different musical genres
Authors: Amandine Pras, Rachel Zimmerman, Daniel Levitin and Catherine Guastavino
Affiliation: McGill University, CIRMMT, Montreal, Quebec, Canada
Published: 127th AES Convention 2009, paper 7879

Abstract: Mp3 compression is commonly used to reduce the size of digital music files but introduces a number of potentially audible artifacts, especially at low bitrates. We investigated whether listeners prefer CD quality to mp3 files at various bitrates (96 kb/s to 320 kb/s), and whether this preference is affected by musical genre. Thirteen trained listeners completed an A/B comparison task judging CD quality and compressed files. Listeners significantly preferred CD quality to mp3 files up to 192 kb/s for all musical genres. In addition, we observed a significant effect of expertise (sound engineers vs. musicians) and musical genres (electric v.s acoustic music).

Documents: slides, paper


Demystifying high resolution audio
Authors: Amandine Pras and Catherine Guastavino
Affiliation: McGill University, CIRMMT, Montreal, Quebec, Canada
Published: 128th AES Convention, May 22-25, 2010, London, UK

Abstract: High resolution audio formats use sampling rates higher than the 44.1 kHz CD standard. But can people hear the difference? In the CIRMMT critical listening room, trained listeners were able to differentiate between orchestral music files at 44.1kHz and 88.2Hz. Except for the sampling rates, the exact same audio gear and settings were used for recording and playback.

Documents: forthcoming

The conclusion drawn by experts were:-
Conclusion
 Trained listeners can hear differences between
CD quality and mp3 compression (96-192 kb/s)
and prefer CD quality.
 Trained listeners can not discriminate between
CD quality and mp3 compression (256-320 kb/s)
while expert listeners could.
 Ability to discriminate depends on listeners’
expertise and musical genre
 Artifacts can be verbalized and do not depend
on musical genre
127th AES Convention, New York City, October 2009.

You can access the copyrighted papers from AES website for fee. You can see the McGill University's slide show http://www.music.mcgill.ca/~hockman/...tation2009.pdf

These papers may not reflect the position of BBC or Harbeth but these are the facts as it stands now. I hope the moderator will let this last post of mine go through because it is based on accepted standards of proper research and for the benefits of users who may able to hear the difference and very much confused like I have been.

ST

[A.S.]

It's easy to quote other people's work, which on the face of what you state above, looks unsurprising, to be expected (that listeners prefer WAV to data-reduced MP3 especially at low bitrates) and rational. So I'm mystified why you believe my, our or the BBC's opinion about the audibility of low bitrate MP3 may be in any way different to that stated above. On the face of it, we're all in agreement. As I've explained in detail, it's a waste of your time focusing on the deep notes of the organ. As I said - the ear is spectacularly poor at resolving low frequencies. So poor that you could play the organ from a 78rpm gramophone disk and it may well sound perfectly OK.

You caught our attention here on the Harbeth manufacturers-run forum with the hint of an interesting DIY listening test which we thoroughly agree is in the spirit of what we are trying to achieve i.e rationalism. My point was that such a test - under the observation of the entire readership and now part of this living archive - needs lots of thinking time and little doing time. The paper you mention above clearly was the result of a carefully constructed series of tests. It is most unlikely that we, as a group, could find (or would want to find) fault with its findings. But that is not to say we can't use that rational approach as a template for how to conduct our own mini-tests.

We here are just not clear what point you are trying to make. Our position going forward is this: we have many templates for how to consider, control and eliminate bias in listening tests. It needs planning. If airtime is to be devoted to personal comparative tests - amplifiers, electronics, stands or audio data-reduction systems, we should follow best practice as you have so conveniently shown us in the above post. That's the standard my Moderators expect to see. That any test has been thought through to the best of our budget and limited ability. Or it's not to be admitted here as part of the archive as future researches will not benefit from it.

P.S. The chart of the frequency span of various instruments is just one snapshot of their sonic capabilities and character. Alone, it's a painfully inadequate one because it doesn't show you how loud the constituent frequencies are within its range. And this whole debate about data-reduction (and how MP3 works) is intimately linked to masking, (not primarily frequency response) and masking is all about how loud sounds mask quieter ones. So that fact that an organ, theoretically, can produce some sort of wheezy whistle at supersonic frequencies just detectable if you stand 1m from the pipe is not shown by the chart. Nor is the dynamic range of the instrument i.e. the range between its maximum and minimum loudness. Both loudness and dynamic range are vitally important factors when we are considering listening tests on a loudness-based (MP3) reduction system and selection of the most challenging instruments for the MP3 system needs much thought. Many interested co-contributors are willing to respond to open-ended questions to help construct valid tests.

[Labarum]

I think, Alan, your views on the simplicity of organ sounds is in need of more investigation.

What you say applies to the basic diapason stops - wood or metal flutes - they have a "pure" sound with relatively few harmonics. I am not sure reed stops and chiff-flutes are in the same class. These may well have quite a lot of energy going way up into the supersonic.

There is an article here http://www.soundonsound.com/sos/aug03/articles/synthsecrets.htm from Sound on Sound about the difficulties of synthesising pan pipes which point to this - there are synthesised components up to 20kHz.

The breathiness of the pan pipes is the "chiff". Some organs have chiff-flutes. If such a chiff-flute were used for the solo line of an organ piece I am guessing switching off the tweeter would change the sound of the music dramatically. The chiff-flute would loose its chiff and its charm.

But I haven't done the test, and I don't have a suitable demo track, but a track of pan-pipes would be a good starting point. The electronic organ makers would know exactly what is needed to add convincing chiff, and would also know how rich are the harmonics of an ear splitting trumpet or tuba stop.

[A.S.]

Since pan pipes obviously don't extend down to subsonic frequencies, they could not have been the organ type shown on the chart which surely was a (metal) pipe organ. So please let's not introduce another level of confusion.

Surely I'm not the only one here who can make a spectral and dynamic range evaluation of a pipe organ? Do we have to rely on other folks charts and reports when it's a simple matter to gather our own data, data adequate for us to move the dialogue forward even if not strong enough for us to publish it in a learned paper? The answer to that question is core to this group.

[Labarum]

Alan, my suggestion is separate and simple: that there are high frequency components in some organ pipes (reeds and chiff-flutes); and that the character of those pipes will not be heard if the high frequencies are filtered out.

[A.S.]

...the character of those pipes will not be heard if the high frequencies are filtered out.

Possibly or probably true. But not definitely true. Surely we're discussing masking here and masking, as stated previously, is one sound swamping another into inaudibility. It is not possible to draw the conclusion you have unless you define the loudness of those high frequencies and all the other sounds occurring at the same instant.

An example to illustrate masking and your pan pipes. A solo pan piper is playing. Then he gives way to full orchestra but continues playing. His pipes are completely masked by the orchestra. It is - and this is the logic behind MP3 encoding - totally irrelevant how rich in harmonics or not his pipes are; it's their individual loudness relative to the other instruments (more accurately, relative to their loudness/frequency spectra) that decides whether or not the pan pipe is passed through MP3 encoding partially or not at all i.e. fully masked. In my example, the sound of the pipes would be dumped as the orchestra dominates the overall loudness. It's conceivable depending on what other instruments are playing along with the pipes and their individual loudness/frequency spectra, that the MP3 coder would eliminate all the pan pipe harmonics just retaining the lower notes, and yes, in those circumstances the pan pipe could come through processing sounding like a regular church organ .... but we'll never know because by the very nature of the MP3 encoding it is impossible to hear how it treats individual instruments in isolation: it works by interpreting and processing the whole sound it is given instant by instant, evaluating that snapshot for loudness in the various spectral bands.

May I remind you that we are drifting away from discussing masking and how MP3 works. If you want to discuss other issues they should be addressed separately from this thread for the benefit of future readers. Let's keep threads on-topic as best we can please.

[Labarum]

Well I thought I was on-message. I suggested that if a stop (a rank of pipes) was used as a "solo instrument" playing the tune and a selection of other stops controlled from another manual were used for the accompaniment, a chiff-flute or reed so used would (by the organists choice) be prominent and constantly unmasked. In such circumstances an organ would test the principles being here discussed.

{Mod comment - thank you but please remember that English is not the first language of many / most Visitors. Also, many would have no understanding of the construction or playing of an pipe (church) organ which is linked to Western religious not eastern music}

[A.S.]

Noted.

We have to find a way to talk reasonably technically about issues like masking (which are fundamentally important in appreciating how we judge sound) without using so many words. I'm cursing myself that I didn't make a TechTalk two weeks ago because I could have demonstrated masking at work using just household props. It would have saved so much typing. And confusion. And hopefully encouraged further debate. But I did ask for feedback and received two comments. Two from over a thousand members. That's not very impressive is it. Sadly, I just do not want to nor do I have the time nor qualifications to cast myself as a seven day a week tutor. Yes, these subjects are hugely important. But we should be adding to and simplifying not merely regurgitating other people's work - that's what makes this forum different. We must apply our own intellect and experience to verify their findings (as best we can) as we have a duty to "educate" and "simplify" using as few, precise technical words as possible. To get to the core of the issue. Using annotated pictures. And sounds. Interactively. To test for ourselves whether we are being sold a story as a fact. To act as a conduit to other less technical readers.

Every passing day I see more clearly how much damage a generation of marketing BS has done. Reasoning and scrutiny has gone out of the window. Do I care? I care because two generations of audio engineers fought tooth and nail to drag up the standards of audio reproduction, and in half a generation that struggle has been abandoned and reversed. That pains me. And in another two generations? All the things we here consider precious will be forgotten.

This group draws by far the most clicks in discussions about amplifiers, cables, stands and all those things which need to be talked about, and where no input is required or expected from Harbeth UK. We have confused our role as a talking shop with that of an 'educator' (in quotes) which consumes a huge amount of time this side. So, going forward, I will conceive of and make TechTalks as and when I find subjects of interest and try to make them as complete and self-contained as possible. Others are most welcome to do so. As a manufacturer's forum we have a legal and moral responsibility to keep 'hard' subjects (those not in the sandbox) sharply in-focus and on-topic and that's where leadership is needed from this side, and the time is regrettably not available to keep a watching brief on too many threads.

We're loudspeaker manufacturers not educators! This thread is closed until I can conceive of a new method to present information succinctly and within our overall Forum Moderation Policy detailed here.