The next-gen HD Audio Bitrate Cheat Sheet!

[Ruined]

A few I will fill in the blanks later

Some quick terms:
Lossy - Some audible information is lost. i.e. rolling off frequencies over 15khz.

Perceptually Lossless - No information audible to humans is lost, however waveform visually looks different than the master. i.e. rolling off frequencies over 19khz, because in repeated scientific tests it was proved that humans can't hear sounds higher than 19khz in frequency.

Mathematically Lossless - No audible information is lost, waveform looks identical to the master. i.e. entire frequency range intact

Bit-resolution/Bit-depth: Wikipedia i.e. 16bit audio. Greater bit-resolution brings greater Signal-to-Noise Ratio.

Sampling Rate: Wikipedia i.e. 48khz audio. For our purposes, divide sampling rate in half to get the max frequency response for a given soundtrack. i.e. 48khz sampling rate = up to 24khz frequency response.

Most common bitrates/configurations:

Dolby Digital Plus (Lossy, though can be Perceptually Lossless at 1.5mbps+ bitrates as per Microsoft)
Bitrates:
16bit/48khz 5.1: 640kbps CBR
24bit/48khz 5.1: 1.5mbps CBR
16bit/48khz 7.1: ?
24bit/48khz 7.1: ?

Dolby Digital TrueHD (Mathematically Lossless)
Bitrates:
16bit/48khz 5.1: 1.5mbps ABR (3.0mbps peak)
24bit/48khz 5.1: 3.4mbps ABR (5.0mbps peak)
16bit/48khz 7.1: 1.9mbps ABR (3.8mbps peak)
24bit/48khz 7.1: 4.7mbps ABR (6.6mbps peak)

DTS-HD (Lossy, though can be Perceptually Lossless at 1.5mbps+ bitrates as per DTS)
16bit/48khz 5.1 (Core): 768kbps CBR
24bit/48khz 5.1 (Core): 1.5mbps CBR
24bit/96khz 5.1: 3mbps CBR
16bit/48khz 7.1: ?
24bit/48khz 7.1: ?
24bit/96khz 7.1: ?

DTS-HD Master Audio (Mathematically Lossless)
16bit/48khz 5.1: ?
24bit/48khz 5.1: ?
24bit/96khz 5.1: 5.3mbps ABR (6.8mbps peak)
16bit/48khz 7.1: ?
24bit/48khz 7.1: ?
24bit/96khz 7.1: ?

PCM (Uncompressed)
16bit/48khz 5.1: 4.6mbps
24bit/48khz 5.1: 6.9mbps
24bit/96khz 5.1: 13.8mbps
16bit/48khz 7.1: 6.1mbps
24bit/48khz 7.1: 9.2mbps
24bit/96khz 7.1: 18.4mbps

[chlngr1970]

Good post Ruined. Thanks for the info...

j

[Taxi]

I didn't buy this when you posted it before, here. I still don't buy it.

As soon as you talk about modifying the source before the compression happens, then you basically invalidate any serious comparison between compression algorithms, lossy or lossless.

[Ruined]

Originally Posted by Taxi

As soon as you talk about modifying the source before the compression happens, then you basically invalidate any serious comparison between compression algorithms, lossy or lossless.

Huh? This is the essence of how lossy encoders work. They throw away frequencies before the "compression" happens to avoid artifacting. With advanced encoders like LAME, you can even adjust what point the threshold should be set at where HF rolloff begins with a commandline parameter. You might not know that its happening, but its generally the first thing that happens when you use a lossy audio encoder. If the frequencies thrown away are ONLY the ones that we cannot physically hear, then WTF is the difference? High frequencies are very difficult to compress without artifacting and sucking up bitrate, so it would be pointless to keep high freqs we cannot even hear in the mix when bitrate is an issue.

I completely do not understand the second half of your sentence.

For reference, with the MP3 lossy codec, here is a table of the default

frequency rolloffs based on bitrate using LAME MP3 encoder:
bitrate (kbps), lowpass (hz)
8, 2000
16, 3700
24, 3900
32, 5500
40, 7000
48, 7500
56, 10000
64, 11000
80, 13500
96, 15300
112, 16000
128, 17500
160, 18000
192, 19500
224, 20000
256, 20500
320, 21000

And, Dolby Digital 5.1:
bitrate (kbps), lowpass (hz)
384, 18000
448, 20000

Finally, DTS 5.1:
bitrate (kbps), lowpass (hz)
768, 19000
1536, none

So if you were use LAME to encode an MP3 at 128kbps, the first thing it would do is rolloff all frequencies above 17.5khz. If it did NOT do any rolloff at 128kbps, say, the result would sound horrible because there is not enough bitrate there to sustain the full frequency range. You'd get full response up to 22hz, but it would be horribly mangled by artifacts (you can test this yourself with commandline parameters). Since those artifacts are much more noticable than the lack of frequencies greater than 17.5khz, the latter is the obvious choice to make. With higher bitrates like 192kbps, we get into territory where the frequencies thrown away do not matter because we can't hear them anyway.

Rereading the second part of your sentence again, I'm not sure you realize that with virtually every lossy encoder known to man the "source" is "modified" pre-compression via a lowpass filter by the encoder, yes (sometimes other techniques, too).

[Taxi]

Originally Posted by Ruined

Rereading the second part of your sentence again, I'm not sure you realize that with virtually every lossy encoder known to man the "source" is "modified" pre-compression via a lowpass filter by the encoder, yes (sometimes other techniques, too).

Yes of course lossy codecs do that, but they're lossy so nobody expects to get all the same bits back after the decode. That's one method they can use to get the encoded file size so much smaller than lossless. But it wouldn't make any sense, at least from an audio quality standpoint, to do that with a lossless encoder.

Anyway, my point is you define terms at the beginning of your post, but I don't think those defintions are too accurate. To say "Lossy - Some audible information is lost. i.e. rolling off frequencies over 15khz." doesn't really get to the heart of what lossy encoding is. "Perceptually lossless" vs. "Mathematically lossless" seems to me to be an attempt at promoting some kind of adgenda regarding these formats.

Lastly, do you have links to any of these scientific studies that prove humans can't hear above 19k? I've never heard that before. (No pun intended. ) From what I've learned from music production, I can tell you that there is a lot more we don't know about psychoacoustics, than what we do know. (Both at the high and low frequency ranges.)

[Derb]

Not quite sure what to make of this thread. Interesting to see the comparative bit rates & such regarding 5.1 & 7.1 lossy & lossless though other than that, everything else is ones opinion over another in terms of what is actually heard.

Sinse I did hear a difference between 2 lossless tracks (16bit/48khz 5.1: 1.5mbps ABR (3.0mbps peak)) & (16bit/48khz 5.1: 4.6mbps) on the same title this statement is void,

Mathematically Lossless - No audible information is lost, waveform looks identical to the master. i.e. entire frequency range intact

[ganthc]

I know for sure that video-wise, I can't always see the imperfections that others see. For example, dugpa finds compressions artifacts all over movies that I just don't see problems with. But with audio, I am much more picky. It has been my main criticism for hd-dvd, which is that from an audio perspective, it is weaker. While colors are often more bright, the PQ really isn't heads above bd. However, bd blows hd-dvd out of the water in the audio department. Hands down, no contest.

The bit rates data is interesting to see though. I'm just not sure how it helps the argument that lossy is better than lossless.

[Ruined]

Originally Posted by Derb

Sinse I did hear a difference between 2 lossless tracks (16bit/48khz 5.1: 1.5mbps ABR (3.0mbps peak)) & (16bit/48khz 5.1: 4.6mbps) on the same title this statement is void,

Mathematically Lossless - No audible information is lost, waveform looks identical to the master. i.e. entire frequency range intact

Your informal listening test does not make my statement "void." TrueHD as a codec is mathematically lossless to the source. That means it is identical, plain and simple. Not much else to say here.

Therefore, if you heard a difference, it is due to one of the following:
1) The mix/master used for the TrueHD and PCM sources was different.
2) One of the soundtracks plays back by default at a louder volume due to Dialog Normalization, and humans tend to associate louder volumes with being better sounding.
3) Somewhere in your processing chain one of the decoders/processors is doing a better job on one soundtrack vs. the other.
4) Placebo effect.

[Ruined]

Originally Posted by Taxi

Yes of course lossy codecs do that, but they're lossy so nobody expects to get all the same bits back after the decode. That's one method they can use to get the encoded file size so much smaller than lossless. But it wouldn't make any sense, at least from an audio quality standpoint, to do that with a lossless encoder.

Again, you are speaking of an encoder that is mathematically lossless. This is an encoder which compresses the file but throws away nothing so that it is mathematically identical to the original. Of course then there would be no lowpass.

Perceptually lossless, though, is something that can be achieved by lossy encoders at high bitrates. It means that some information was thrown away from the original signal, but the information thrown away was inaudible and thus is not really "lost" because it was not able to be heard in the first place. Good MP3 encoders like LAME tend to reach perceptual lossless at around 256kbps, WMA reaches it 192kbps VBR, DD+ and DTS both are claimed to reach it at 1.5mbps. While you are not getting the original information bit-for-bit, you are getting all the information audible to humans bit-for-bit. Hence, it is perceptually lossless but not mathematically lossless.

Anyway, my point is you define terms at the beginning of your post, but I don't think those defintions are too accurate. To say "Lossy - Some audible information is lost. i.e. rolling off frequencies over 15khz." doesn't really get to the heart of what lossy encoding is.

I thought it was a very simple and easy to understand example. Lossy encoding at low bitrate does throw away audible frequencies, such as 15khz+. At high bitrates, though, the frequencies tossed are those that we cannot hear anyway, hence the difference between perceptual lossless and low bitrate lossy.

"Perceptually lossless" vs. "Mathematically lossless" seems to me to be an attempt at promoting some kind of adgenda regarding these formats.

My agenda is to help educate people on a topic that seems to be misunderstood. It is worthwhile to know the differences between the different types of encodings available and their results. For instance, technically a 24bit/48khz 1.5mbps DD+ track should be perceptually lossless - meaning you shouldn't be able to tell the difference between this track and an uncompressed 24bit/48khz PCM track. Take that a step further, if you only have 1.5mbps of headroom to encode, is it better to have a mathematically lossless 16bit track or a perceptually lossless 24bit track? With the former you get a carbon copy of the master, but it is a master that has been downsampled and is at a lower bit resolution than the original. With the latter you don't get a carbon copy of the master, but you should get all of the audible information and at a higher bit resolution no less.

That is where the interesting discussion takes place, and audio engineers appear to be divided on this topic.

Lastly, do you have links to any of these scientific studies that prove humans can't hear above 19k? I've never heard that before. (No pun intended. ) From what I've learned from music production, I can tell you that there is a lot more we don't know about psychoacoustics, than what we do know. (Both at the high and low frequency ranges.)

The Nyquist Theorem. Also, there are a ton of tests that indicate on the hydrogenaudio forums that 19-20khz+ is inaudible to humans, and people cannot tell the difference between an uncompressed CD and a 256kbps MP3/192kbps WMA - which both happen to roll off frequencies above 19-20khz+ depending on the parameters.

A number of links indicating that we can't hear past 19-20khz:
http://everything2.com/index.pl?node_id=812113
http://en.wikipedia.org/wiki/Nyquist...mpling_theorem
http://www.pcmus.com/digthe.htm#brick
http://csunix1.lvc.edu/~snyder/2ch11.html

The sampling rate of CD was specifically picked because the frequency range provided by the sampling rate was just above the threshold of human hearing.

[tomdkat]

Originally Posted by Taxi

"Perceptually lossless" vs. "Mathematically lossless" seems to me to be an attempt at promoting some kind of adgenda regarding these formats.

He isn't making those terms up or anything. I did a search on Wikipedia (not that it's the definitive source) on WMA Lossless and ran into those terms as well.

Lastly, do you have links to any of these scientific studies that prove humans can't hear above 19k? I've never heard that before. (No pun intended. ) From what I've learned from music production, I can tell you that there is a lot more we don't know about psychoacoustics, than what we do know. (Both at the high and low frequency ranges.)

Well, I believe this is one of the things THX processing does in an A/V receivers. It rolls off frequencies above a certain threshold hold, in part to prevent the audio from being distressing to people when viewing movies in a home environment. I've personally heard this kind of effect when comparing the DD and dts-es discrete audio on Gladiator with THX Cinema mode engaged in my THX Select certified A/V receiver. I don't know what the actual threshold is, but the 19khz Ruined cites sounds like the kind of frequency range I remember reading about in past DD vs dts debates.

Peace...

[tomdkat]

Originally Posted by Derb

Sinse I did hear a difference between 2 lossless tracks (16bit/48khz 5.1: 1.5mbps ABR (3.0mbps peak)) & (16bit/48khz 5.1: 4.6mbps) on the same title this statement is void,

Mathematically Lossless - No audible information is lost, waveform looks identical to the master. i.e. entire frequency range intact

How can one sample of two lossless tracks make the Mathematically Lossless statement void??????

Peace...

[tomdkat]

Originally Posted by Ruined

The Nyquist Theorem. Also, there are a ton of tests that indicate on the hydrogenaudio forums that 19-20khz+ is inaudible to humans, and people cannot tell the difference between an uncompressed CD and a 256kbps MP3/192kbps WMA - which both happen to roll off frequencies above 19-20khz+ depending on the parameters.

I'm not challenging the Nyquist Theorem but I've run into an exception of the example you cite above about CD audio. I increased the sampling rate of iTunes from the default of 128 kbps to 320 kbps because the iPod owner was dissatisfied with the audio quality of the music she ripped from CDs she owns. I increased the bitrate gradually, first starting at 192 kbps and then 256 kbps. As I increased the bitrate, I would play the original CD version first and then the ripped version. When I increased the bitrate to 320 kbps, she thought the ripped version sounded better than the original CD.

Others I've talked with who own iPods are happy with the default sampling rate, which does lend itself to support your example above.

I guess the bottom line is it really depends on the sensitivity of the ears of the listener.

Peace...

[Taxi]

Originally Posted by tomdkat

He isn't making those terms up or anything. I did a search on Wikipedia (not that it's the definitive source) on WMA Lossless and ran into those terms as well.

Oh, I wasn't attributing any adgenda to Ruined specifically. (Sorry, I guess it does read that way.)

Just like in the divx days when the terms "Standard DVD" and "Open DVD" were used to try to imply different things about DVD, so too could terms like "mathematically lossless" and "perceptually lossless" be used.

Lossless is just lossless. There's no need to qualify it. Just as there is no need to prefer it over the different lossy codecs. They can all do a good job when done well.

[tomdkat]

Originally Posted by Taxi

Oh, I wasn't attributing any adgenda to Ruined specifically. (Sorry, I guess it does read that way.)

Ok.

Lossless is just lossless. There's no need to qualify it.

I dunno, I think there is a need to qualify it but I do agree that these technical details aren't required knowledge by the CE consumers nor will these details make any difference in the practical application of them in this context (audio on HD movie releases).

I think a lot of people want lossless audio for "just because" reasons, which is fine, but they attempt to justify their desire with technical debate and then the cookie starts to crumble.

Peace...

[ganthc]

And I think that Taxi has said that the human ear and hearing is not something that necessarily fits neatly into the techno-jargon bit rate schema. Every individual would probably have different hearing abilities, much like fingerprints. Some people would not be able to discern the difference between CD and itunes downloads. I certainly can. Some people can't pick out an oboe from an orchestra playing either. But to blanket say that DD+ 1.5mbps is equal to lossless is false. True, we don't have comparisons from the same source, but from what I have heard of them, lossless still sounds better, more robust, and has a better immersion of sound.

Audio equipment and setups can also play a part in this. All I am saying is that with my setup, equipment, and hearing, I can't ever say that lossy audio sounds equal to or in the same vicinity as lossless. I'm not saying that lossy sounds like crap. But when you are in a highdef format, I want the best sound that goes with the best picture. And it seems that with hd-dvd, they aren't delivering the best sound possible 8 times out of 10. The sound is good, but it's not the best.

[Ruined]

Originally Posted by ganthc

Some people would not be able to discern the difference between CD and itunes downloads. I certainly can.

Well now you are talking about 128kbps, not 192kbps or 256kbps. Big difference, and no one ever claimed 128kbps AAC was perceptually lossless as its far from it. On the other hand, 256kbps AAC likely sounds no different than the original CD and hence would be considered perceptually lossless at that bitrate.

Some people can't pick out an oboe from an orchestra playing either. But to blanket say that DD+ 1.5mbps is equal to lossless is false. True, we don't have comparisons from the same source, but from what I have heard of them, lossless still sounds better, more robust, and has a better immersion of sound.

Right, we can't test every individual in the world but listening test after listening test shows the same trend - at some point high bitrate lossy encoding sounds just as good as the uncompressed source.

Audio equipment and setups can also play a part in this. All I am saying is that with my setup, equipment, and hearing, I can't ever say that lossy audio sounds equal to or in the same vicinity as lossless. I'm not saying that lossy sounds like crap. But when you are in a highdef format, I want the best sound that goes with the best picture. And it seems that with hd-dvd, they aren't delivering the best sound possible 8 times out of 10. The sound is good, but it's not the best.

Ah, but this is where the division comes in. If a DD+ 1.5mbps 24bit/48khz *is* in fact perceptually lossless as it is claimed to be, then technically you cannot state it would be worse than say a lossless 16bit/48khz PCM track. Reason being that although the lossy track does not have all the original PCM track's information, it actually would have more accurate audible information since the PCM track's bit resolution was downsampled to 16bit. Further, you said you want the best picture - a 24bit/48khz DD+ 1.5mbps track takes up less than half the space of an equivalent 24bit/48khz TrueHD track and more than 3x less peak bandwidth. Even against a 16bit/48khz TrueHD track, though the same amount of space is taken up the DD+ track has half the peak bandwidth.

So in conclusion, the debate is if in fact DD+ 1.5mbps is perceptually lossless as claimed, are you not getting the same or better audio quality in less space/bandwidth - which can be used for interactive extras, additional soundtracks, higher video bitrate, or other features?

Anyway, it is quite an interesting discussion and there appears to be no definite answer to it.

P.S. - You can also use imaging to help with this association. With Photoshop, when you save your image in the lossy format JPEG, you can pick varying levels of compression. If you set it to maximum quality, usually you can't tell the difference between the source and the JPEG. You still aren't getting the original, but you are getting enough visual data that it looks the same as the original. With lossy audio encoders, its a similar situation.

[tomdkat]

Originally Posted by ganthc

And I think that Taxi has said that the human ear and hearing is not something that necessarily fits neatly into the techno-jargon bit rate schema.

I think it does since there are things that can be scientifically proven, like frequencies the human ear can actually hear. It's just the technical jargon doesn't really matter since in the real world you're going to choose the audio you want to listen to given the options available to you. The technical jargon comes into play when people try to explain their preferences in a technical manner vs just being happy with their preference.

Every individual would probably have different hearing abilities, much like fingerprints.

I would say that WILL have different hearing abilities as well as preferences to what sounds "good". To some, simply sounding very LOUD is translating to sounding "good" while that won't be the case to someone else.

But to blanket say that DD+ 1.5mbps is equal to lossless is false. True, we don't have comparisons from the same source, but from what I have heard of them, lossless still sounds better, more robust, and has a better immersion of sound.

Well, that would depend on the analysis used. If looking purely and objectively at the raw performance capabilities of DD+ 1.5MBps and LPCM or Dolby TrueHD (or whatever else lossless audio you want to consider), it might be possible to prove DD+ 1.5Mbps performs equally. The thing is, numbers don't or can't describe how something will sound to any given person. A test can prove a set of Bose speakers can produce as much sound as B&W speakers but that test can't prove everyone will consider both sets of speakers as sounding the same when playing the same audio.

Audio equipment and setups can also play a part in this. All I am saying is that with my setup, equipment, and hearing, I can't ever say that lossy audio sounds equal to or in the same vicinity as lossless. I'm not saying that lossy sounds like crap. But when you are in a highdef format, I want the best sound that goes with the best picture. And it seems that with hd-dvd, they aren't delivering the best sound possible 8 times out of 10. The sound is good, but it's not the best.

Audio equipment certainly does play a part in this and an important part at that. Still, what you describe here is simply personal preference. If I were to go to your place and listen to what you consider to be an awesome LPCM audio track on some Blu-Ray release (say Pirates of the Caribbean), would I think the same? Would I find it too loud or too soft? Would I find it overwhelming or underwhelming? Hence the "problem" with using vague terms like "best".

Peace...

[tomdkat]

Originally Posted by Ruined

Ah, but this is where the division comes in. If a DD+ 1.5mbps 24bit/48khz *is* in fact perceptually lossless as it is claimed to be, then technically you cannot state it would be worse than say a lossless 16bit/48khz PCM track. Reason being that although the lossy track does not have all the original PCM track's information, it actually would have more accurate audible information since the PCM track's bit resolution was downsampled to 16bit.

This touches on something else I think is VERY important but is being overlooked. What is it we're actually hearing? If the DD+ audio is perceptually lossless, that means it will sound just like the actually lossless audio even though not all the bits are there. If we could do blind audio tests and find most of the people couldn't tell the difference between actual lossless audio and perceptually lossless audio, then Ruined's assertion is correct even though the actual lossless audio still might be preferred. If the blind audio test proves most COULD tell the difference between actual lossless and perceptual lossless, then Ruined needs to apologize to us all for wasting our time.

It's not the bits that are in the audio that's important, it's how the audio sounds after reproduction by our A/V gear that's important. When you hear a great audio track you comment on how good or great the audio sounds and don't comment on how happy you are all those bits are present. Comments like, "Wow, that kicked my ass!" apply solely to how the audio sounds pure and simple.

I think it would be great to head over to Dolby Labs and do some audio tests between DD+, Dolby TrueHD, and LPCM audio.

Peace...

[Ruined]

Originally Posted by tomdkat

If we could do blind audio tests and find most of the people couldn't tell the difference between actual lossless audio and perceptually lossless audio, then Ruined's assertion is correct even though the actual lossless audio still might be preferred. If the blind audio test proves most COULD tell the difference between actual lossless and perceptual lossless, then Ruined needs to apologize to us all for wasting our time.

Well, like I've said before these blind tests have been done with MP3, WMA, and AAC stereo audio codecs. And in all cases, it was found that at some bitrate the audio becomes "perceptually lossless" (usually around 192kbps-256kbps depending on the codec) despite being a "lossy" codec. Many claim that they can definitely hear a difference, but once you put them in a blind test situation and crank up the bitrate to 256kbps with a quality encoder (LAME or Windows Media Encoder) they are unable to differentiate - most who make this claim are probably thinking of 128kbps encodes, shoddy defunct encoders like XingMP3, or simply used to shoddy amateur encodes off Limewire.

No tests have been done with DD+ AFAIK due to the newness of the codec combined with the lack of access to encoding tools and the fact that we are just now getting access to uncompressed/lossless 5.1 masters. My guess, though, is that you will see the same thing with DD+ as you do with other "lossy" codecs - at some point, you reach a state of "perceptual lossless" sound, which is claimed to be 1.5mbps for DD+.

Believe me, I understand the fascination with lossless audio. I was in the process of archiving my CDs with lossless WMA. Now I'm having second thoughts. Why waste all that space with 50-60mb files when its proven 6-7mb files sound the same? Those bigger files mean that I need a clunkier MP3 player since 8gb won't cut it for a library of lossless files. Not to mention the smaller files are much more portable.

[Taxi]

Originally Posted by Ruined

And in all cases, it was found that at some bitrate the audio becomes "perceptually lossless" (usually around 192kbps-256kbps depending on the codec) despite being a "lossy" codec.

I guess the part I have a problem with is the term "perceptually lossless". That's like saying that my Pontiac is a "perceptual Ferrari" at certain speeds. (Like, say, under 10 miles an hour. )

Lossy isn't lossless. If you want to say the the audio is "perceptually identical" to lossless, that is one thing. But "perceptually lossless" sounds like more of a political statement to me, and therefore I distrust its use. (Maybe I should substitutle "political" with "marketing". Equal distrust there on my part.) I think that people (not Ruined in particular) who use "perceptionally lossless" are trying to avoid saying "perceptually identical" because they believe that claim is harder to support.

Maybe it *is* time to call the folks at Audio F/X. Seamonkey? Are you up for it?