View Single Post
Old 30th January 2002, 18:57   #4  |  Link
tangent
Ogg + Vorbis + XviD
 
tangent's Avatar
 
Join Date: Dec 2001
Location: Singapore
Posts: 182
Quote:
Originally posted by ChristianHJW

Make sure you use bitrates > 160 kbps for your MP3 and set the 'Stereo' mode in Lame ( thats the MP3 encoder of BeSweet ) to 'Stereo' instead of 'Joint Stereo' to avoid the encoder does 'channel coupling', because this could destroy the DPL matrix and lead to sonic artefacts on center and surround speaker.

A good mode to set Lame is VBR q = 1 .... 3 , i always use 'old routine' , its slow but very good. Dont use the 'R3MIX' preset because this one uses 'Joint Stereo' by default. Its excellent for music or if you wnat a normal Stereo movie track ( no DPL ), but not for surround.
I have to disagree here. If you understand how Dolby Surround and Mid-Side stereo coding works, you will realise that Lame's Joint Stereo makes encoding of Dolby Surround encodings much more efficient. How Dolby Surround basically works is that rear channel information is encoded into the left and right channel with a 180 degrees phase shift. The easiest way to describe it is that you place the sample into the left channel, and place the inverted version of the sample into the right channel. During decoding, the difference between the left and right channel (L-R) is taken to obtain the rear channel sample.

There are two types of stereo coding modes used in MP3. They are intensity and Mid-Side stereo. Intensity stereo is meant to be used only for low bitrates because it kills phase information. LAME does not implement Intensity stereo. Mid-Side stereo frames are used by LAME when Joint Stereo is enabled. Mid-Side stereo works by converting the left and right channels into mid and side channels. Mid=(L+R)/2 and Side=(L-R)/2. The idea is that there is usually a lot information which are similar between the left and right channels, therefore a near mono sample window will probably have most of the information in the mid channel and nearly nothing in the side channel. This allows the information in the mid channel to be coded with higher quality because more bits are available to encode them.

Now look what happens with a Dolby Surround sample is being encoded: Redundant data actually appears, but this time not in the Mid channel but in the Side channel! In fact, the Side channel is now a perfect representation of the Dobly Surround rear channel! As you can see, Dolby Surround encodes is likely to have better quality using Joint Stereo than simple stereo. It's almost like MP3 Mid-Side stereo was designed with Dolby Surround in mind

The main worry about badly implemented joint stereo is the loss of stereo seperation and a bad JS/SS threshold, and not really the loss of Dolby Surround information. Joint Stereo does not use 100% MS frames, but a decision is made whether to use a MS or SS frame by looking at the amount of redundant data. If the left and right channel are very dissimilar (occurs more often in music), then it would revert to a SS frame. Normal VBR is known to have some issues with this decision process, but this is fixed by using the --nssafejoint switch together with the --nspsytune switch. With this combination, you will never have to worry about joint stereo problems, and yet benefit from the more efficient coding of joint stereo. --r3mix does not use this --nssafejoint, but it is used in "--alt-preset standard" and "--alt-preset extreme". The ABR --alt-presets are designed to use this only from 160kbps onwards.
tangent is offline   Reply With Quote