In the previous post I asked at the end: ‘Is 24 bits it worth it?’ I want to do an extensive post on this in due course, but let me first make my point briefly. In that previous post I contrasted the 2032kbps average bitrate of the Australian (and UK, I learn) version of Slumdog Millionaire against the 3962kbps of the US version. Our version is 16 bits; their version is 24 bits.
I point out that to go up from 16 to 24 bit uncompressed PCM requires a 50% increase in space. But this increase in DTS-HD Master Audio requires a 95% increase in space. The nice thing about Slumdog Millionaire is that it is likely identical sound, aside from the bit depth, making it highly comparable. But to confirm, I have eleven other Blu-ray movies with English language 5.1 channel DTS-HD Master Audio sound with 16 bits of resolution. Their average bitrates range from 1834kbps to 2504kbps, and the average of the eleven is 2114kbps. I have four Blu-ray movies with English language 5.1 DTS-HD Master Audio sound with 24 bits of resolution. They range from 3849 to 4530kbps and average 4253kbps.
That figure is twice (2.01x) the average bitrate for the 16 bit movies, so that basically confirms the drop-off in efficiency with the compression algorithm when moving from 16 bit to 24 bit sound.
Double check and repeat the process with the Dolby TrueHD tracks I have scanned (omitting Gandhi and Gigi, because they are both pre-5.1 channel days even though they are rendered this way on disc). I have fourteen 5.1 channel movies with TrueHD in 16 bits and these range from 1376kbps to 2172kbps, and average 1668kbps. The seven equivalents in 24 bits range from 2535 to 3525 and average 3109kbps. That’s an 86% increase in size.
This has led me to think that there is actually limited value in using 24 bit sound.
In brief: the efficiency of lossless compression systems like DTS-HD MA and Dolby TrueHD depends, amongst other things, on the predictability of the data. The severe drop-off in efficiency when you go from 16 to 24 bits suggests that most of those extra eight bits are unpredictable. The most likely reason for the unpredictability is that their values are largely random. Random samples are noise, pure and simple. White noise to be precise.
Looking at it from another angle, the noise floor of the microphones used to record stuff is typically -96dB or higher, and actual (rather than nominal) signal to noise of much equipment would rarely be more than 96dB. The 17th bit on a 24 bit scale is at -96dB. It isn’t surprising that the bottom eight bits of 24 bit sound should mostly be noise.