# 192 kbs and 320 kbs, is there really a difference?

Discussion in 'Sound Science' started by clincher09, Sep 3, 2008.

1. Quote:

 Originally Posted by jung /img/forum/go_quote.gif I want VBR 400kbps, not 200kbps, nor lossless, and never CBR. I don't care how many people can not ABX. If only one person in the whole world can ABX, it is one too many. But lossless is only a mathematical illusion, not a physical reality. The real physical world is quantum and random. There is no such thing as an exact physical copy. So VBR with high enough bitrate, maybe 400 to 500 kbps, will be physically the same as the original. Not an approximation, not just not ABX'able, but the same, in physical sense.

Aren't you contradicting yourself here? You say there is no such thing as an exact physical copy, and in the next sentence you claim that 400kbs VBR is physically the same as the original? Make up your mind

2. Quote:

 Originally Posted by hybris /img/forum/go_quote.gif Aren't you contradicting yourself here? You say there is no such thing as an exact physical copy, and in the next sentence you claim that 400kbs VBR is physically the same as the original? Make up your mind

High bit rate lossy compression can be exactly the same as lossless, in the physical sense, not just in the sense of whether there is audible difference.

The physical performance is the target. Lossless compression is an approximation of that physical reality. A high bit rate lossy compression can be no worse than lossless, not farther away from the performance.

When digitized, dithering must be applied. The best dithering is random in nature, like the quantum randomness of the real physical world. So there are countless different ways of arranging the bits for dithering. These differently dithered bits will not be the same bit for bit, but they are equal representations of the physical reality. So bits can be changed, by high bit rate lossy compression, but the result can be equal to lossless in the above physical sense.

Another example, a performance is recorded in HiFi analog. Then it is digitized into two sample rates, 44/16, and 96/24. The 44/16 is then lossless compressed to 900kbps. 96/24 is lossy compressed to 500kbps. The lossy 500kbps 96/24 can be closer to the analog master than the lossless 900kbps 44/16.

3. Contributor
Quote:

 Originally Posted by jung /img/forum/go_quote.gif High bit rate lossy compression can be exactly the same as lossless, in the physical sense, not just in the sense of whether there is audible difference.

No, since there is data loss they cannot be the same physically. They may be the same perceptually or even functionally but not physically.

Quote:

 The physical performance is the target. Lossless compression is an approximation of that physical reality. A high bit rate lossy compression can be no worse than lossless, not farther away from the performance.

A lossless compression is not physically the same as an uncompressed file but when uncompressed the data is bit for bit identical, arguing whether they are in fact the same bits is missing the point

Quote:

 When digitized, dithering must be applied. The best dithering is random in nature, like the quantum randomness of the real physical world. So there are countless different ways of arranging the bits for dithering. These differently dithered bits will not be the same bit for bit, but they are equal representations of the physical reality. So bits can be changed, by high bit rate lossy compression, but the result can be equal to lossless in the above physical sense.

No , dithering always operates on the low bits, you trade off some extra non correlated noise against quantization error, and so while the distribution is random there are finite limits to what changes can happen.

Also, the conversion to lossless or lossy compression operates after all this has happened. Dithering is not relevant at that point unless you are also downsampling as well which is not the same thing as just lossy vs lossless compression

Quote:

 Another example, a performance is recorded in HiFi analog. Then it is digitized into two sample rates, 44/16, and 96/24. The 44/16 is then lossless compressed to 900kbps. 96/24 is lossy compressed to 500kbps. The lossy 500kbps 96/24 can be closer to the analog master than the lossless 900kbps 44/16.

Exactly wrong. With the first case as long as you sample at fs2 and have a dynamic range of < 96db 16/44.1 will capture all the information you need to accurately recreate the waveform. However you do not get any more actual information in the 24/96 but you then slash its data content by a factor of 9.

Even mathematically you are wrong. A 24/96 recording has 4608000 bits per second and a 16/44.1 has 1411200 bits per second. The 24/96 has 3.26 times more data , however all the data from the 16/44.1 is retained when losslessly compressed, for the 24/96 file almost 90% of the data is thrown away , any really high frequencies caught by 24/96 gone - too expensive to retain, then there is the peceptual coding aspect this includes two forms of masking one removes frequencies that are too low to be heard above the dominant ferquencies and the temporal masking removes proximal stimuli , both of which fundamentally alter the waveform. In short the lossy compression must be less accurate than lossless compression.

4. Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif No, since there is data loss they cannot be the same physically. They may be the same perceptually or even functionally but not physically. In short the lossy compression must be less accurate than lossless compression.

First let's define "lossy", so that we are talking about the same thing. I think most people agree that "lossy" means after the codec, the bits are not completely identical to lossless. So lossy does not necessarily mean that data are lost, but just bits are different.

Then back to dithering. As I said, dithering is absolutely necessary. And the best dithering is random. So after dithering, there is not just one bit arrangement that is the best, but there are numerous different bit arrangements that are the best, as good as, equal to, all the other in the same group.

So let's look at two different dithered result, A, and B, both from the best dithering method, but with different dithered bit results. So A and B are equal, in the absolute physical sense, but they are different bit for bit.

Then let's assume we apply a high bit rate lossy codec on A. The result is not bit identical to A. But for some reason let's assume it is identical to B. So the lossy codec can produce a result that is physically equal to lossless, but not bit to bit identical.

Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif Exactly wrong. With the first case as long as you sample at fs2 and have a dynamic range of < 96db 16/44.1 will capture all the information you need to accurately recreate the waveform. However you do not get any more actual information in the 24/96 but you then slash its data content by a factor of 9.

We are interested in the absolute physical reality. The analog master has information beyond the frequency and quantization limit of digitization. So it needs to be filtered during both 96/24 and 44/16. But for 96/24 less is filtered out than for 44/16. So 96/24 is closer to the analog master than 44/16.

Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif Even mathematically you are wrong. A 24/96 recording has 4608000 bits per second and a 16/44.1 has 1411200 bits per second. The 24/96 has 3.26 times more data , however all the data from the 16/44.1 is retained when losslessly compressed, for the 24/96 file almost 90% of the data is thrown away

You would not say that lossless throw away 30% of data, why would you say that lossy throw away 90%?

Let's modify the example to show it more clearly. Let's say the lossless 44/16 result in 900kbps. Now let's assume a lossy 96/24 results in 900kbps, too. With a good lossy codec, the 900kbps 96/24 must have more information than the lossless 900kbps 44/16. A good lossy 900kbps 96/24 must be closer to the analog master than a lossless 900kbps 44/16.

5. Contributor
Quote:

 Originally Posted by jung /img/forum/go_quote.gif . So lossy does not necessarily mean that data are lost, but just bits are different.

You misunderstand. Lossy compression throws away data it has to !, the codec decides what data can be imperceptibly thrown away and then throws it away, like really high frequencies and frequencies that cannot be heard de to masking. It is not like a zip file where the data is simply re-coded in a more efficient form, data is lost by definiton, that is why it is called lossy, and once gone you can never he it back.

Quote:

 So let's look at two different dithered result, A, and B, both from the best dithering method, but with different dithered bit results. So A and B are equal, in the absolute physical sense, but they are different bit for bit.

No, they are not the same in the absolute physical sense if the bits are not the same they are not the same, they may not be audibly dfferent but that is not the same thing at all.

Quote:

 Then let's assume we apply a high bit rate lossy codec on A. The result is not bit identical to A. But for some reason let's assume it is identical to B. So the lossy codec can produce a result that is physically equal to lossless, but not bit to bit identical.

You really are not grasping the concept of lossy encoding nor of what it means to be physically identical.

To be identical in any meaningful way two digital files have to terminate with identical bit patterns, they can be different but perceptually indistinguishable , but that is not the same thing.

Quote:

 We are interested in the absolute physical reality. The analog master has information beyond the frequency and quantization limit of digitization. So it needs to be filtered during both 96/24 and 44/16. But for 96/24 less is filtered out than for 44/16. So 96/24 is closer to the analog master than 44/16.

In absolute terms if there is real frequency information above 22khz then yes that is so.

Quote:

 You would not say that lossless throw away 30% of data, why would you say that lossy throw away 90%?

Because it does. Lossless throws away nothing it merely re-encodes it in a more space efficient manner. Lossy does throw away data, it leaves sufficient data to recreate a waveform that is more or less perceptually identical to the original but you go from ~ 1.14mbps to 245kbps or whatever level you choose , by any interpretation data is lost. Now if you had said musical information that would be closer to the mark, but words are important.

Quote:

 Let's modify the example to show it more clearly. Let's say the lossless 44/16 result in 900kbps. Now let's assume a lossy 96/24 results in 900kbps, too. With a good lossy codec, the 900kbps 96/24 must have more information than the lossless 900kbps 44/16. A good lossy 900kbps 96/24 must be closer to the analog master than a lossless 900kbps 44/16.

Try this, let us hypothesize that our analog souce has frequencies up to 200khz and a dynamic range of 50 bits and two channels. So we have a datastream of 40Mbits/second - so the 16/44.1 captures 2.85% of it and the 24/96 captures 11.52% so far so good - now a lossless encode of this throws away nothing so we still have 2.85%. However the lossy encode of the 24/96
turns 4608000 bps into 900000bps so the final data is in fact 1.28% of the original.

6. lossy means data is lost. hence the name...

7. Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif You misunderstand. Lossy compression throws away data it has to !, the codec decides what data can be imperceptibly thrown away and then throws it away, like really high frequencies and frequencies that cannot be heard de to masking. It is not like a zip file where the data is simply re-coded in a more efficient form, data is lost by definiton, that is why it is called lossy, and once gone you can never he it back.

If we can not agree on what is lossy codec, as opposed to lossless, then we can not have this discussion meaningfully.

My point is that bits can be different, but the result can still be equal, physically. See below.

Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif No, they are not the same in the absolute physical sense if the bits are not the same they are not the same, they may not be audibly dfferent but that is not the same thing at all. You really are not grasping the concept of lossy encoding nor of what it means to be physically identical. To be identical in any meaningful way two digital files have to terminate with identical bit patterns, they can be different but perceptually indistinguishable , but that is not the same thing.

Let me clarify the terms I use.

Mathematically identical, means that all bits are identical. The normal sense of lossless codec.

Physically equal, means that two digital files are equally related to the original physical reality, e.g., the analog master. The two digital files can be not bit identical, but still equally related to the physical analog master, using the best scientific instruments available, not just human perception.

E.g., different dithered results, A and B, are physically equal, even though not bit identical. B does not contain less information than A, nor does A than B. A and B are physically equal.

Then after a lossy codec on A, suppose the result happens to be bit identical to B, but not bit identical to A. Since B is physically equal to A, so the lossy codec does not throw away any information, 0%.

Quote:

 Originally Posted by nick_charles /img/forum/go_quote.gif Because it does. Lossless throws away nothing it merely re-encodes it in a more space efficient manner. Lossy does throw away data, it leaves sufficient data to recreate a waveform that is more or less perceptually identical to the original but you go from ~ 1.14mbps to 245kbps or whatever level you choose , by any interpretation data is lost. Now if you had said musical information that would be closer to the mark, but words are important. Try this, let us hypothesize that our analog souce has frequencies up to 200khz and a dynamic range of 50 bits and two channels. So we have a datastream of 40Mbits/second - so the 16/44.1 captures 2.85% of it and the 24/96 captures 11.52% so far so good - now a lossless encode of this throws away nothing so we still have 2.85%. However the lossy encode of the 24/96 turns 4608000 bps into 900000bps so the final data is in fact 1.28% of the original.

A 44/16 wave file has 1411kbps of bits, but not 1411kbps of real data information. If it does, then lossless compression is not possible. Suppose a lossless codec can compress it down to 900kbps, then the original wave must have less than 900kbps of data information. Among the original 1411kbps, lots of bits are redundant, not real data information. So if a lossy codec compresses it down to 900kbps, it can be keeping all 100% of the original data information, throwing away 0%, exactly the same as the lossless codec. We don't say that it throw away 30% of real data information just because it compresses 1411kbps down to 900kbps.

8. Quote:

 Originally Posted by jung /img/forum/go_quote.gif E.g., different dithered results, A and B, are physically equal, even though not bit identical. B does not contain less information than A, nor does A than B. A and B are physically equal.

Yes. If A and B contain sufficient information to trace back to the master. The only way I can see this happen is that A and B are both lossless.

For example:

The master source = XXXXXXXXXYYYYYYYYY (18 slots)

A = 9*XYYYYYYYYY (12 slots)
B = XXXXXXXXX9*Y (12 slots)

Then A and B are equivalent.

Quote:

 A 44/16 wave file has 1411kbps of bits, but not 1411kbps of real data information. If it does, then lossless compression is not possible. Suppose a lossless codec can compress it down to 900kbps, then the original wave must have less than 900kbps of data information. Among the original 1411kbps, lots of bits are redundant, not real data information. So if a lossy codec compresses it down to 900kbps, it can be keeping all 100% of the original data information, throwing away 0%, exactly the same as the lossless codec. We don't say that it throw away 30% of real data information just because it compresses 1411kbps down to 900kbps.

Now let's use that example.

Suppose the 96/24 is the master, i.e = XXXXXXXXXYYYYYYYYYZZZZZZZZZ (27 slots)

A 44/16 sample = XXXXXXXXXYYYYYYYYY (18 slots)

A lossless resulted from the 44/16 sample: A = 9*XYYYYYYYYY (12 slots)

A lossy resulted from the 96/24 master: B = XXXXYYYYZZZZ (12 slots)

Suppose that [Z] is useless information, then it's obvious that A is a better file and B (Suppose that there is absolutely no perceivable difference between 44/16 and 96/24).

9. Quote:

 Originally Posted by jung /img/forum/go_quote.gif If we can not agree on what is lossy codec, as opposed to lossless, then we can not have this discussion meaningfully. My point is that bits can be different, but the result can still be equal, physically. See below.

It's hard to agree when you obviously have no idea how lossy encoding works. And I must say it's really hard to grasp what you mean with this "physically equal" term you keep throwing around. The closest I can get is that you appear to think that if it sounds the same it's "physically equal". That's most likely not true. Your ears are not an instrument of absolute fidelity.

Quote:

 Originally Posted by jung /img/forum/go_quote.gif Let me clarify the terms I use. Mathematically identical, means that all bits are identical. The normal sense of lossless codec. Physically equal, means that two digital files are equally related to the original physical reality, e.g., the analog master. The two digital files can be not bit identical, but still equally related to the physical analog master, using the best scientific instruments available, not just human perception. E.g., different dithered results, A and B, are physically equal, even though not bit identical. B does not contain less information than A, nor does A than B. A and B are physically equal. Then after a lossy codec on A, suppose the result happens to be bit identical to B, but not bit identical to A. Since B is physically equal to A, so the lossy codec does not throw away any information, 0%.

A lossy encoded file can and will never be "equally related" to the physical original analog master, and not with a file that's just dithered either. If it is, it's a really poor lossy encoder, and will hardly compress at all. Based on what you are saying it appears that you think a lossy encoder doesn't do much except dithering (how else could a lossy file be equal to "B" which is a dithered original audio stream).

That is not the case.

Quote:

 Originally Posted by jung /img/forum/go_quote.gif A 44/16 wave file has 1411kbps of bits, but not 1411kbps of real data information. If it does, then lossless compression is not possible. Suppose a lossless codec can compress it down to 900kbps, then the original wave must have less than 900kbps of data information. Among the original 1411kbps, lots of bits are redundant, not real data information. So if a lossy codec compresses it down to 900kbps, it can be keeping all 100% of the original data information, throwing away 0%, exactly the same as the lossless codec. We don't say that it throw away 30% of real data information just because it compresses 1411kbps down to 900kbps.

That is theoretically correct, but a lossy encoder that compresses a wave file to 900kbps most likely WILL NOT retain all the physical data(that was never the design goal of the encoder), and slightly beside the point you will find the fact that is essentially stupid to use a lossy encoder if you don't want to compress more than what a lossless encoder can achieve. Why? Because it will remove data, while you could have avoided that using a lossless( = removes no data) encoder. It is correct that it will not remove 30% of actual (useful) audio data, but it will not be lossless.

I must say I have no idea whatsoever what you are trying to say here. Lossy encoding will never result in a "physically identical" (whatever that means) result as the original. It may (and often will) sound the same, and that's what matters. But it's not physically the same, as DATA IS LOST.

10. Contributor
Quote:

 Originally Posted by jung /img/forum/go_quote.gif If we can not agree on what is lossy codec, as opposed to lossless, then we can not have this discussion meaningfully.

Everyone knows what the terms mean, but you seem to be trying to re-define English.

Lossy = Data has been lost
Lossless = No data has been lost

I suspect that you've seen lower bit rates on lossless files compared to the original and assumed that it means data has been lost, when the bitrate is only the average rate of data per second. While compressed, it's obviously lower as less data has to be sent to result in the same amount of information.

Quote:

 My point is that bits can be different, but the result can still be equal, physically. See below.

This is impossible, period. Bits are different = different data. You're saying that 1 can equal 0 or 0 can equal 1.

Quote:

 Let me clarify the terms I use. Mathematically identical, means that all bits are identical. The normal sense of lossless codec. Physically equal, means that two digital files are equally related to the original physical reality, e.g., the analog master. The two digital files can be not bit identical, but still equally related to the physical analog master, using the best scientific instruments available, not just human perception.

Since data is lost, with ANY lossy compression, the analogue output after decoding will result in sounds (detail) lost. It is impossible for it to be otherwise. Lossy and lossless digital can't be defined relative to the analogue original. By its very nature, digital encoding removes part of the analogue signal. You're confusing the issue trying to define things in terms of the analogue original.

Quote:

 E.g., different dithered results, A and B, are physically equal, even though not bit identical. B does not contain less information than A, nor does A than B. A and B are physically equal. Then after a lossy codec on A, suppose the result happens to be bit identical to B, but not bit identical to A. Since B is physically equal to A, so the lossy codec does not throw away any information, 0%.

Both are impossible. Both do not happen, period. Not in this universe anyway.

11. 320 mp3 earnt my respect when I failed a blind test with with lossless using hd595 and yamaha receiver. Different story with 650 and xcan valve amp. First to go is the full body midrange under complex passages.

192 is unworthy of audiophile status

12. Contributor
Quote:

 Originally Posted by jung /img/forum/go_quote.gif If we can not agree on what is lossy codec, as opposed to lossless, then we can not have this discussion meaningfully.

Its hard to agree when you clearly have no idea how lossy and lossless encoders work.

As 'Currawong' say:
Lossy = Data has been lost
Lossless = No data has been lost

13. 320KBS with full stereo and 44.1 KHZ and so on, can sound really good!
But
Still the lead goes for the lossless format like FLAC.

Beside, 320KBS take a lot of space, and so does FLAC, but FLAC is better, so why will I use 320KBS ?!

As much as I see it, it is 192KBS - 256KBS in VBR MP3 (or what ever VBR format) when someone don't have enough space ans still want some kind of quality.

And full LOSSLESS format like FLAC when you don't have space problems (like most of us now days) and want to know you have a full BIT PERFECT quality (as the original CD)

14. I have a severe space problem so I can't enjoy flac... I love flac dearly but my main source of music is a 20gb mp3 player with optical out so I just use 192kb vbr ogg. The reason I use an mp3 player as source is that bill gates has given me too many hours of misery trying to make the computer digital output sound as good as my mp3 player's. I actually prefer 192 vbr out of my mp3 player than flac out of my current computer usb software's output. And now that mbd2884 and others using the same gear as me found out that usb-audio makes the computer sound much better, I am still unable to use it because no matter how much I try/crash I can't get it to work on vista -_-. I have a theory that the bad economy is entirely due to "lost work time" due to vista.

15. Why is there such huge debate over lossy vs lossless in this thread when the question is over 192 and 320, which both are lossy?

But yeah it's a pity Haloxt you are unable to get it to work, cause the only reason I would use the ASIO4All is if I had no available ASIO drivers to use. Good luck!