Originally Posted by clancy688
First of all, thanks for all answers, especially to amirm! Your post was most helpful.
You are welcome. It used to be what I did for a living (managing development of audio technology including lossy and lossless compression).
Is this simply *one* channel? Therefore 2 * 160 kbit/s AC3 = 320 kbit/s would be right on the threshold of having a perceptible difference to the source?
No. AC-3 like most other lossy codecs uses two methods to save bits in stereo/multi-channel encodings;
1. Shared buffering. There is a single buffer pool (data rate to be used) that each channel dips into. Think of it as having a family meal at a restaurant with a party of 2 or more. If one channel is more difficult to encode than the other, it will use more bits than the other. This pays huge dividends in multi-channel coding where most of the channels are idle or playing at low levels. Think of the center channel carrying most of the audio signal in a movie track. AC-3 will then allocate most of its bandwidth to that channel, resulting in far higher bit rate and much reduced distortion than as if it statically allocated bandwidth to each channel.
2. Channel coupling (also called Joint Stereo in other codecs). High frequencies, especially those that have transients, are very difficult for lossy codecs to encode. Depending on bit rate and number of channels, AC-3 encodes frequencies above certain level (theoretically 3.5 KHz but practically 10 Khz) in mono and then sends a signal to the decoder to play these at the levels of the original signal. The idea being that your ear is more sensitive to the envelop of the sound (level) more than its spectrum make up. So if you make sure that you get the relative levels right, the fact that you have a mono signal won't be noticed. Or at least, it won't be noticed as much as compression artifacts. Needless to say, you can't pull this trick for mono signals.
Those 320 kbps clips are stereo, I assume?
The test was 384 kbps and not 320. All tracks were encoded as 5.1. However, some of them used to be stereo and were converted to multi-channel using signal processing.
Thanks, I think I'll try this. But it may take some time as I've not much spare time available right now. But I know someone researching audio coding at Fraunhofer IIS (they created MP3 and were involved in AAC), so I hope I can get those MPEG reference clips you're mentioning.
The names of them were in the 384 kbps test graph I showed. But yes, your FHG contact most definitely has it. It is like a hammer to a carpenter
. He may also have more direct info on this topic. I used o have an AC-3 encoder years ago and did a bunch of tests with it but no longer do these days.