320kbit: Blade vs. Lame, 44.1Khz and 48.0Khz Sampling Options

[Ultrasound]

I am new to these boards, but not new to quality recordings. In fact, as my name implies, I work with sounds in the kilo-hertz, mega-hertz, and tera-hertz. I'm a tweaker for sure.

I did a comparision of some 320kbit recordings using two different codecs - I was not able to compare to FhG, but at 320kbit, I'm sure it sounds pretty good. Here goes my layman's evaluation:

dBPowerAmp Music Converter - Release 10.1
http://www.dbpoweramp.com/

Rio Cali MP3 Player - 768MB
http://www.digitalnetworksna.com/shop/_tem...o.asp?model=258

Sennheiser HD 497 Headphones
http://www.sennheiserusa.com/newsite/pdfs/hd497.pdf

Codecs: Blade 0.94.2, Lame 3.96.1
Methods: 320kbit CBR stereo, 320kbit CBR joint-stereo, 256kbit VBR Stereo, 256kbit Stereo @ 44.1Khz Sampling Rate and 320kbit CBR Stereo @ 48.0Khz Sampling Rate

Media: Great White - Call it Rock and Roll; Evanescence - Bring Me to Life; Evanescence - Everybody's Fool

SCOPE:

The purpose of the listening test was to see if there were any discernable differences (to my ears) between the Blade codec and the Lame codec at 320kbit Stereo. Suffice it to say, I could not really hear a difference at that quality level. I went down as low as 256kbit and still could not really detect a difference between the two.

OVERVIEW:

I understand the masking and joint-stereo techniques, but I would still rather have true stereo - Fraunhofer states " ... joint-stereo is used in cases where only low bitrates are available but stereo signals are desired." Disk-space and file-size are of no consequence to me. I am after the best quality for lossy encoding.

Perhaps the differences of the two codecs become more apparent at 192kbit or less ... I could not hear a difference between 320kbit CBR or 320/256kbit VBR either.

The biggest difference of all, and most noticeable, was when sampling at a higher frequency. I did not see many posts concerning sampling at 48Khz but please offer your opinions.

EVALUATIONS - Great White:

This rock band is from the 1980's and has some killer crispness and transient responces in their recordings. The track was Call it Rock and Roll. Try as I might, listening for the top-hat cymbols and snar-drums did not reveal any differences between the Blade and the Lame codecs at 320kbit.

When I sampled the track at 48,000Hz the difference was immediate. The Blade encoder sounded more 'spacious' ... the Lame codec also was pleasingly spacious but sounded a bit 'tighter'.

EVALUATIONS - Evanesence:

A relatively new band, this singer is a female and they know how to rock. They incorporate piano, strings, along with the amplified components. With the track Bring me to Life, the same findings hold true with the two codecs at 320kbit.

Moving on to the sampling rate change yielded more surprises. Once again, the Blade 'opened up', as did the Lame - both were very much an improvement on sound-staging when compared to the 44.1Khz sampling rate. This time, I was able to put something on the Blade at 48Khz ... the opening piano sequence that sounded more spacious, was in fact what I believe to be distortion. The Lame was also more spacious at this sampling rate, but it retained it's composure and 'tightess'. I did not notice any distortion with the Lame codec at 48Khz samplings.

I added one more track with some nice light strings at the beginning - Everybody's Fool. Bit-rates above 256k did not tell a story, but the sampling rate did again. In this track, both codecs sounded slightly more open than at 44.1Khz, but the Blade had a bit more ... perhaps the stories of how many codecs fall off with filtering at around 16,000Hz plays a roll - that is if the Blade does not filter the highs and goes on to say 18,000Hz. This could explain the 'extra' little bit the Blade seemed to have in most cases. This little bit of 'extra' in the Blade codec results in distortion at higher frequencies, however.

CONCLUSIONS:

I created 18 tracks between the 3 sound tracks and various combinations ... here are a few of the particulars:

Great White - Call it Rock and Roll
01 - Blade - 320k - 44.1Khz Sample - Stereo - 48sec encode
02 - Lame - 320k - 44.1Khz Sample - Stereo - 47sec encode
03 - Lame - 320k/256k VBR - 44.1Khz Sample - Stereo - 1min 24sec encode
04 - Lame - Insane - 44.1Khz Sample - Joint-Stereo - 47sec encode
09 - Blade - 320k - 48Khz Sample - Stereo - 47sec encode
10 - Lame - 320k - 48Khz Sample - Stereo - 1min 10sec encode

Evanesence - Bring Me to Life
11 - Blade - 320k - 48Khz Sample - Stereo - 39sec encode
12 - Lame - 320k - 48Khz Sample - Stereo - 1min 16sec encode

Evanesence - Everybody's Fool
17 - Blade - 320k - 48Khz Sample - 28sec encode
18 - Lame - 320k - 48Khz Sample - 55sec encode

I may end up going to 256kbit CBR or perhaps a VBR with a high-end cap of 320kbit and low-end cap of 256kbit as shown in track #03. The difference in sound-staging from track #10 to track #02 was evident. I'm not sure what the technical reasoning for that is, but it did make a difference.

Notice track #11 and track #12 ... the Lame took quite a bit more time in processing and it paid off in the results.

The 'distortion' I mentioned before, with the Blade codec at 48Khz, sounded as if things were run through a Peavy amp with a switch thrown between 'distortion' and 'tube' - but left in the 'distortion' mode. Maybe it is that the Blade goes a tad bit higher in frequency response and does not handle it well at 48Khz - maybe not. In any event, the Blade codec did not handle 48Khz very well - what was intitially thought to be 'spaciousness' turned out to be what I would characterize as noise, or distortion. No matter what I gave to the Lame codec, it never mis-behaved. The highs were crisp and with 48Khz, the sound-stage got wider (but just by a little).

I'm not sure if I will re-sample my tunes at 48Khz at this point - I'll need to play around with it some more. It might be fun for you to try some as well and see what you get.

I'd have to agree: Lame 3.96.1 is a great sounding codec - though not decisive at 320kbit, it does seem to be the codec to choose.

[sven_Bent]

I understand the masking and joint-stereo techniques, but I would still rather have true stereo - Fraunhofer states " ... joint-stereo is used in cases where only low bitrates are available but stereo signals are desired." Disk-space and file-size are of no consequence to me. I am after the best quality for lossy encoding

Read the FAQ

[Hanky]

I work with sounds in the kilo-hertz, mega-hertz, and tera-hertz.

That's great for you, but unfortunately we don't.

[outscape]

are your results based on blind listening tests? i still don't understand why you insist on using blade. it's an old, crappy mp3 encoder that is not in development for 2 years and has more than its share of flaws.

[shadowking]

I cannot understand this whole blade thing. Try music with sharp drum attacks and there is no need to even abx.

To Ultrasound: Try abx and you will find at bitrates over 128k things start to 'even out' with modern codecs. You won't find find many humans that can pick out well tuned codecs at 192k with normal music.

[rjamorim]

...and tera-hertz.

Yayyyy, man. 1,000,000,000,000Hz. Way to go.

[Yaztromo]

...and tera-hertz.

Yayyyy, man. 1,000,000,000,000Hz. Way to go.

lol

Please read HA terms of service Ultrasound. Espcially TOS 8.

http://www.hydrogenaudio.org/forums/index.php?showtopic=3974

[westgroveg]

At this place you are required to provide double blind listening test proof when making quality claims. You may hear sharper, tighter, and flatter differences but unless you can prove you did, it doesn’t help much.

[cabbagerat]

...and tera-hertz.

Yayyyy, man. 1,000,000,000,000Hz. Way to go.

Considering that the atomic radius of nitrogen is 68e-12 metres and sound travels at approximately 330metres per second in air, the wavelength of a terahertz sound wave is only slightly larger than a single N2 molecule.

It's time to start marketing a 2,000,000,000kHz/16 format - for those important sub-molecular harmonics.

[Gabriel]

Let me repeat the opinion of other people, in a different way:

You seems to be very interested in reaching the optimal sound quality. I would advise you to learn about ABX tests. If you take the time to learn about it, I am sure you will not regret it. Things will be different to you after, you will be very surprised.
To learn about ABX, I think that you are on the right website.

BTW: I would advise you let Lame choosing itself the stereo encoding method.

[SirGrey]

I understand the masking and joint-stereo techniques, but I would still rather have true stereo - Fraunhofer states " ... joint-stereo is used in cases where only low bitrates are available but stereo signals are desired." Disk-space and file-size are of no consequence to me. I am after the best quality for lossy encoding

>>I understand the masking and joint-stereo techniques
No, you don't.
To encoding be at best quality you require mid-side stereo to be used.

See here for explanations: http://www.hydrogenaudio.org/forums/index....showtopic=24029

[Benjamin Lebsanf...]

do you mean this ?
http://www.hydrogenaudio.org/forums/index....t=ST&f=15&t=995

[Gecko]

As for 48kHz sampling: Lame is mostly tuned for 44.1kHz. Don't know about Blade. It probably isn't tuned much for anything. It's basically just the iso reference code.

Resampling to 48kHz from 44.1kHz will decrease sound quality (slightly). If done properly, there will be no higher frequency in the 48kHz file than the Nyquist frequency of the 44.1kHz file (which would be 22.05kHz). You are not increasing temporal resolution or something.

Back in the r3mix.net days, some tried resampling to 48kHz for other reasons: to decrease temporal smearing a little when encoding to mp3. I don't remember the results, but I assume it was not worthwhile or we would be using it today.

The doctor who checked my sinuses recently used ca. 50kHz. Note that it is very difficult to properly emit frequencies this high from a resonating component into air. Most of the energy doesn't enter the air but is reflected back at the boundaries of the emitting medium. That's why the doctor uses gel. Bats go a long way to get this done.

OT: Here's something fun to try: take your super platinum edition pro plus tweeter, run 50kHz through it and point it at flies. They will think your tweeter is a bat and as a reflex drop to the ground. Other insects like moths start flying spirals.

[Ultrasound]

Thanks for those who attempted to offer something ... as I stated, this was just my interpretation of those listening tests.

As for joint-stereo, one person said to " Read the FAQ", which I have done, but all I did was quote verbatem from Fraunhofer's site - it says that joint-stereo is for when you have lower bit-rates. When you compare two channels of stereo, throw out some 'redundant' bits to save space, you are losing something.

When I said I understand the techniques used, another person said "No, you don't.
To encoding be at best quality you require mid-side stereo to be used." JS is used to compress the file size - you have to get rid of something. Typically this encodes lower frequencies in mono while higher frequencies are still stereo. Intensity stereo ... as for mid-side stereo, I'm not a fan of creating a middle channel (L+R) and then a side of (L-R) ... I'd rather stay as true to the original as possible. How can one rationalize that to get the "best" sound quality, joint-stereo is "needed" as opposed to the original stereo channels? Joint-stereo and M/S are techniques used to reduce file size - Frauhofer is correct: if you want smaller file size (lower bitrate) then this is one technique used to achive that. If you are not concerned with smaller file size, then stick to stereo and don't tinker with the phase information. The original signal is stereo ... you force the algorithm to change that.

Yes, I work with kilo-hertz and on up to tera-hertz ... someone had to comment on that too. I use FFT - I've seen educated talk here in these forums - but I hoped to get better reponses from your experience with MP3s. I use gel and water, like one person commented in here - ultrasound does not travel well in air. Although I am new to these boards, I do know something about sound waves - I'm a scientist/engineer with NASA doing working with them everyday.

The "whole thing about Blade" was in an effort to answer some posts that said nothing had been done around the 320kbit range. I stated how I perceived that Blade was noisy and had distortion, and in the end, Lame was the best - no need to get out of shape thinking you have a Blade supporter or whatever.

I do not have the programs some of you have, but I will try to get them. I did, to the best of my ability, try to convey my interpretations - if this is not good enough to meet TOS 8, then I am sorry. My equipment at work will not work with MP3 files. I thought some people here might be friendly enough to objectively receive some feedback on a personal test - I am beginning to wonder.

I do plan to learn about the ABX programs you all speak of ... I'm sure it will be rewarding.

Sampling at a higher resolution should improve aliasing artifacts - all I said was that with Lame, it might be fun for some to give it a try. There is a reason newer CDs, DVDs, and professionals are recording at higher sampling rates of 48Khz, 96Khz, etc ... AAC, for example, uses a Modified Discrete Cosine Transform, and samples up to 96Khz. With the 24Khz bandwidth afforded by the 48Khz sampling rate, some of the high frequency roll-off could be accounted for - that was the thinking anyway.

Sorry if I "teed" the ball up too high and some took a swing. Does anyone have anything constructive to offer about using Lame with 48Khz sampling? I would like to know if there is anything to it, and if anyone has tried it.

[Ultrasound]

Back in the r3mix.net days, some tried resampling to 48kHz for other reasons: to decrease temporal smearing a little when encoding to mp3. I don't remember the results, but I assume it was not worthwhile or we would be using it today.

Thanks, Geko ... good info.

[boojum]

Does anyone have anything constructive to offer about using Lame with 48Khz sampling?  I would like to know if there is anything to it, and if anyone has tried it.

My general experience is that I do not hear better with my mouth open. YMMV.

[Pio2001]

all I did was quote verbatem from Fraunhofer's site - it says that joint-stereo is for when you have lower bit-rates

Maybe it is true for Fraunhofer's encoders but if it is the case, then Lame is different. And there was that famous Fraunhofer version which had a nasty bug that completely destroyed the stereo imaging when Mid/Side was enabled...

JS is used to compress the file size - you have to get rid of something.

No, you don't get rid of anything when you convert LR to MS, the process is lossless, exept for rounding errors. It has nothing to do with intensity stereo, that is a lossy process for low bitrates.

You miss the point. The L/R to M/S conversion is not part of the psychoacoustic compression. The encoder removes (we should rather say "adds quantization noise") things where they have the smallest audible impact on the sound, but the M/S conversion is not a part of this process.

don't tinker with the phase information.

The phase is unchanged.

M=L+R
S=L-R

Thus,

L=(M+S)/2
R=(M-S)/2

Nothing is lost.


In order to give an idea, the room comes from the fact that you usually convert something like

Left = 123456
Right = 123410

Into

Mid = 246866
Side = 000046

And, that, in a very imaged way, in MP3, you may write 000046 as 46, saving the space taken by the four zeros.
That's not what actually occurs, of course, in reality, both channels are psychoacoustically compressed, and the side channel can be, with the same quality, more compressed than both left and right.
For a given target bitrate, with the room that is thus freed, you increase the bitrate of both Mid and Side channel, which allows to improve the fidelity to the original compared to Left/Right encoding, not to decrease it.

Another thing, Lame does not always use Mid Side, but smartly switches between L/R and M/S according to the most efficient method. That's what is called Joint Stereo.

I thought some people here might be friendly enough to objectively receive some feedback on a personal test - I am beginning to wonder.

That's right, such kind of feedback is not welcome on Hydrogenaudio ! It may sound harsh, but I prefer to state it straight in order to avoid unuseful discussions !
But once you have read a little bit about ABX and blind listening tests, I hope you will understand what I mean. We do appreciate that you came to share the fruit of your efforts, it's just that the results are not useful according to the board's criterions, that are summarized into the TOS number 8, which is very strict. The reason for this is explained here : http://www.hydrogenaudio.org/forums/index....showtopic=11442

I do plan to learn about the ABX programs you all speak of ... I'm sure it will be rewarding.

Maybe this will help. I wrote it recently (the date of the post is wrong). I didn't get any feedback about it yet, but I think that it explains in detail the ABX system : http://www.hydrogenaudio.org/forums/index....showtopic=16295

There is a reason newer CDs, DVDs, and professionals are recording at higher sampling rates of 48Khz, 96Khz, etc ...

This is another can of worms. The lastest thread about it brought very controversial point of views on this matter : http://www.hydrogenaudio.org/forums/index....topic=26218&hl=

[Gecko]

Thanks, Geko ... good info.

Just for clarification, what these people did was:
- rip a CD (44.1kHz)
- resample to 48kHz
- encode to mp3

Resampling looses quality. I don't remember whether the resulting mp3s had less temporal smearing (which would have been the desired effect) or were suffering from new problems. For all I know, it could have been a total fiasco.

The reason that some encoders support "hi-rez" formats is simply that there is no reason not to. Fidelity only plays a very minor roll. Your encoder will apply the same ATH curves and add quantization noise many orders higher than the attainable noise floor of 24bit recordings as it would to your regular 44.1kHz/16bit signal. From the encoder's point of view, the signal it has to encode is very much the same.

[Ultrasound]

Pio2001 & Geko ... thanks for your intelligent replies. I do accept your rules and understand that my experiements are not welcomed, as presented. Sorry if I broke some rule about that.

Can anyone advise if there is a drawback to using stereo with Lame, as opposed to joint-stereo?

[Digisurfer]

Although I am new to these boards, I do know something about sound waves - I'm a scientist/engineer with NASA doing working with them everyday.

As a scientist/engineer you can surely appreciate the usefulness of objective testing rather than subjective. While I must admit that I can't really offer anything particularly useful to this thread, I can confirm that ABX testing, which I finally got around to experimenting with a few days ago using Foobar and the comparator plug-in, has be nothing short of a revelation. Definitely well worth learning and implementing if audio quality is really important to you, and I can now easily see why TOS #8 is so important on Hydrogenaudio.

[saratoga]

Can anyone advise if there is a drawback to using stereo with Lame, as opposed to joint-stereo?

Assumeing you mean full stereo, quality will be reduced and/or file size will mushroom. At 320kbps where you tested, file size cannot be increased to accomidate ineffcient stereo modes, so the encoder is forced to discard additional information.

[SirGrey]

all I did was quote verbatem from Fraunhofer's site - it says that joint-stereo is for when you have lower bit-rates.

Pio2001 already posted a great explanation.
What I feel I must to add - Fhg often used "joint-stereo" term in mean of ms-is stereo (mid-side intensity stereo) which is lossy mode and is used to increase quality on low bitrates. Do not mix it with ms stereo, which is lossless.
I think it is the root of all that misunderstanding about joint-stereo modes...
EDIT:

Can anyone advise if there is a drawback to using stereo with Lame, as opposed to joint-stereo?

Quality should suffer a bit. Lame can efficiently switch between L/R and M/S mode when needed, thus saving a bit space for more info.
IMHO, on such a high bitrate it will be very hard to notice the difference (if possible)

This post has been edited by SirGrey: Aug 27 2004, 23:38

[Pio2001]

Sorry if I broke some rule about that.

No problem. For my part, I always consider it Ok the very first time someone breaks it, because it is a very unusual rule.
Of course, once people get reminded of it, further violations lead to the increasement of the warning level.

[Ultrasound]

Thanks again ... I did manage to obtain a program called "SpectroGram" and compared a time sample from CD, Lame, and Blade. One was nearly like the CD ... you can see the picture - have you guys used that program? I won't post the results, unless you ask, or I can figure out what the procedure is.

ABX is next on my list - I agree on the 48Khz rate, but at 44.1Khz, I was surprised at the SpectroGram.

[audioflex]

There is no sense in upsampling to 48khz (24khz) m8, the CD only has data up to 44.1khz (22khz), 48khz was meant for downsampling from 96khz (48khz), not upsampling from 44khz.