I have been looking into this since I need to losslessly encode several 5.1 surround tracks. None of the lossless codecs available on the internet seem to have any form of multichannel decorrelation, only stereo pair decorrelation with M/S technique.
MLP (as used in DVD-A) has a multichannel decorrelation matrix, which should allow to bring 5.1 tracks down rather much.
Surround recordings generally have alot of information that is common for all channels, since sounds are usually panned into several speakers at once.
I have taken a look at the KLT transform for this purpose, but I don't know if the output from it would be suited for the type of predictors used in FLAC at least.
If anyone has done any research into this topic please tell me and the codec developers about it. I think it is time for the free lossless codecs to go multichannel!!!
Martin Saleteg
N i n j a F X
Full Version: Multichannel decorrelation for lossless codecs
I was under the impression that FLAC supported multichannel audio, although I think that it might be in the format specification only and not actually implemented anywhere.
The design goals of WavPack 4.0 include full support for multichannel audio based on MicroSoft's extended RIFF wav spec:
http://www.microsoft.com/whdc/hwdev/tech/a...multichaud.mspx
However, my plans are to simply encode with a combination of stereo pairs and mono channels. Taking advantage of inter-channel correlation really doesn't buy that much extra compression in stereo (like 1% - 2%, iirc), and I suspect that applying this to 5.1 would result in similar results (and worse if there was delay between channels).
BTW, most of the lossless compressors I am familiar with do a little more that just the MS transform. Monkey's Audio and WavPack use the MS transform and then additionally decorrelate between the mid and side channels. I believe that OptimFROG and LA don't use the MS transform at all but have very sophisticated cross-channel decorrelation techniques (and achieve the best overall performance).
I have actually been looking for 5.1 samples to use at some point in my development and would very much appreciate you contacting me if you could make some available (or anyone else who knows a source of these).
The design goals of WavPack 4.0 include full support for multichannel audio based on MicroSoft's extended RIFF wav spec:
http://www.microsoft.com/whdc/hwdev/tech/a...multichaud.mspx
However, my plans are to simply encode with a combination of stereo pairs and mono channels. Taking advantage of inter-channel correlation really doesn't buy that much extra compression in stereo (like 1% - 2%, iirc), and I suspect that applying this to 5.1 would result in similar results (and worse if there was delay between channels).
BTW, most of the lossless compressors I am familiar with do a little more that just the MS transform. Monkey's Audio and WavPack use the MS transform and then additionally decorrelate between the mid and side channels. I believe that OptimFROG and LA don't use the MS transform at all but have very sophisticated cross-channel decorrelation techniques (and achieve the best overall performance).
I have actually been looking for 5.1 samples to use at some point in my development and would very much appreciate you contacting me if you could make some available (or anyone else who knows a source of these).
QUOTE(bryant @ Jul 27 2003, 08:05 PM)
However, my plans are to simply encode with a combination of stereo pairs and mono channels.
That's exactly what Flac does in multichannel mode. That's why he wants a coder with multichannel decorrelation.
In my tests, I have noticed that MLP doesn't compresses much the files that are fed to it. I believe it's mostly to consume few resources on decoding (for DSP implementations), ancillary data (signature fields to validate the copy, watermark, flags, decoder requirements information...), and error correction/detection code.
Besides, I think it might be optimal for multichannel audio, and I tested it mostly on stereo streams.
I'd say the gain depends on the source material alone. If there are many panned sources (as there usually are in movies and music mixes) you will get quite some benefit from the decorrelation.
I have been experimenting with a bunch of 5.1 sound effects recorded in spaced-cardioid form (using the SPL Atmos system). I don't get much gain using M/S decorrelation, tops 1-2%, and I don't really expect to gain that much more for multichannel decorrelation, due to the amount of time information present using this technique.
This paper describes the MLP and its matrix:
http://www.meridian-audio.com/w_paper/mlp_jap_new.PDF
There is quite some research being done in the lossless area by Dai Yang, and at the upcoming AES convention she will present this paper, which sounds very interesting to me:
D. Yang and T. Moriya, “Lossless compression for audio sources with IEEE floating point format”, to appear in the 115th Convention of Audio Engineering Society, New York, NY, Oct. 10 – 13, 2003.
She has done quite alot of work regarding multichannel decorrelation, exploring the KLT and other transforms for this purpose...
As I don't have an MLP encoder, what is the average bitrate you seem to get out of the encoder for a 5.1 file? I can present you with samples if you want to...
Martin Saleteg
N i n j a F X
I have been experimenting with a bunch of 5.1 sound effects recorded in spaced-cardioid form (using the SPL Atmos system). I don't get much gain using M/S decorrelation, tops 1-2%, and I don't really expect to gain that much more for multichannel decorrelation, due to the amount of time information present using this technique.
This paper describes the MLP and its matrix:
http://www.meridian-audio.com/w_paper/mlp_jap_new.PDF
There is quite some research being done in the lossless area by Dai Yang, and at the upcoming AES convention she will present this paper, which sounds very interesting to me:
D. Yang and T. Moriya, “Lossless compression for audio sources with IEEE floating point format”, to appear in the 115th Convention of Audio Engineering Society, New York, NY, Oct. 10 – 13, 2003.
She has done quite alot of work regarding multichannel decorrelation, exploring the KLT and other transforms for this purpose...
As I don't have an MLP encoder, what is the average bitrate you seem to get out of the encoder for a 5.1 file? I can present you with samples if you want to...
Martin Saleteg
N i n j a F X
QUOTE(rjamorim @ Jul 28 2003, 04:46 AM)
QUOTE(bryant @ Jul 27 2003, 08:05 PM)
However, my plans are to simply encode with a combination of stereo pairs and mono channels.
That's exactly what Flac does in multichannel mode. That's why he wants a coder with multichannel decorrelation.
In my tests, I have noticed that MLP doesn't compresses much the files that are fed to it. I believe it's mostly to consume few resources on decoding (for DSP implementations), ancillary data (signature fields to validate the copy, watermark, flags, decoder requirements information...), and error correction/detection code.
Besides, I think it might be optimal for multichannel audio, and I tested it mostly on stereo streams.
Where do you get the MLP encoder?
QUOTE(Diocletian @ Aug 10 2003, 08:57 PM)
Where do you get the MLP encoder?
http://www.meridian-audio.com/p_mlpenc.htm
Cheapo!
QUOTE
"Price reduction! Due to the rapidly increasing demand for DVD-Audio content tools, we are now able to offer the encoder at $2500." 
Damn!!
Best price for a complete Discwelder Chrome/Surcode MLP encoder package at the moment (feb 2004):
http://www.midi-classics.com/sw/swM6621.htm
Discwelder Chrome/Surcode MLP Encoder
List: 4,490.00
Sale: $3,199.95
For the separate Surcode MLP encoder:
List: 2,495.00
Sale: $1,679.95
That saves you a lot of money compared to the listed prices on the Minnetonka website :-)
By the way in practice a 50% lossless reduction in average size is attainable for MLP encoded 5.1 24/96 surround files, and even a larger reduction for MLP encoded 24/192 stereo files. It means that a combined 5.1 24/96 group together with a 24/96 or 24/192 stereo track easily fits on a standard 4.7 Gbyte DVD-A disc.
Hippo
http://www.midi-classics.com/sw/swM6621.htm
Discwelder Chrome/Surcode MLP Encoder
List: 4,490.00
Sale: $3,199.95
For the separate Surcode MLP encoder:
List: 2,495.00
Sale: $1,679.95
That saves you a lot of money compared to the listed prices on the Minnetonka website :-)
By the way in practice a 50% lossless reduction in average size is attainable for MLP encoded 5.1 24/96 surround files, and even a larger reduction for MLP encoded 24/192 stereo files. It means that a combined 5.1 24/96 group together with a 24/96 or 24/192 stereo track easily fits on a standard 4.7 Gbyte DVD-A disc.
Hippo
That's That's THAT's Amazi- . . . . . Oh wait, I really don't care.
Just here to bitch at your for resurrecting a long dead thread.
Peace
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.