Like you say, the difference is essentially one of trust. I 'trust' 16bit uncompressed audio more because I know exactly what I'm discarding when I dither and truncate the 24bit file (usually done in the one step), so I can reasonably assure myself that nothing is missing from the range of data that contains the 'musical information'. I know that the only thing I'm discarding is the bottom 8-bits, the part that is nothing but noise.
If I know the dynamic range of the musical information in my 24bit recording, I can make sure that it will fit into the range of my 16bit target file with all of the musical information intact.* Often I needn't do anything to make sure that information fits because most recordings have a program dynamic range that is considerably less than 16bits. For some extremely dynamic things I've recorded, using the 24bit recording format to best advantage with lowish levels that left plenty of extra headroom to accommodate unexpected peaks, I might need to raise the program level and its associated noise floor somewhat prior to the dither&truncate step to make sure all the musical information fits comfortably in the 90+ dB range of the 16bit target file without the quietest events, buried deep in the recorded noise floor, dropping below the dither/truncate level. In doing that I'm just shifting the 'musical information range' upwards a few bits, trading some completely unused bits at the top of the range for a few extra at the bottom, which I know are probably nothing but noise anyway. I'm just insuring that the musical information range of the data is completely unchanged by the 24bit > 16 bit file format change.
With perceptual encoding, I'm always left wondering just what was discarded. Even if I don't hear a difference, I can't completely trust that something barely perceptible in the 'musical information' range of data wasn't thrown out. Actually I know for a fact that data was discarded from that range. But I have to trust that the codec made wise enough choices that I won't notice that missing information.
*I love 24bit as a recording format because it lets me capture a dynamic range considerably larger than that of what I'm recording. It means I don't have to be as much of a soothsayer in determining perfect level settings. But I'm totally fine with 16bits as a completely sufficient target format. That's because I have total control over the levels at that point and I know that all the musical information is retained intact.
I still listen to and play 24bit files primarily, but that's mainly about convenience- that's the raw file I have and it takes more effort to make a quality 16bit file from it. Yet I don't think 24bit files are an inherently better delivery format for any musical recording who's dynamics fit comfortably within 16bits, which is most everything musical and anything that can be played on a home system (exception for rocket liftoffs and cannon shot recordings with crickets in the background perhaps). I'm not saying that no differences can ever be heard between playback of otherwise musically identical 16 and 24 bit files, but I think those differences, if detectable, are likely due to DAC technology, not inherent in the file format itself, and will likely continue to become less and less significant as DAC quality continues to improve.
For similar reasons I use simple triangular dither, simply because I know exactly what it is doing and I’ve found its known contribution to be inaudible. I’ve done the dither listening tests and I can only hear a difference if I increase the levels to a completely un-natural and unusable volume which for normal music would rip your head off and which no system could ever play. This is especially the case for most everything I’ve recorded, even symphonic music and eclectic threshold of hearing stuff never has a super low-level, studio quiet noise floor. I’m comfortable that the audible difference between triangular and noise shaped dithers is less than I’ll ever hear at any normal level. By extension, I’m comfortable that I’m not loosing any potentially audible information with a 16bit target delivery file format. I can’t say the same for perceptual coding like mp3. Noise shaped dither (also perceptually based) shifts the added noise around to less audible regions which might affect other things I do with the file even if I can’t hear that noise. In that way those dither schemes are similar to the mp3 unknowns.
The perceptual technologies are fascinating, but they are essentially human hearing tricks. Of course I use and enjoy mp3 as a lightweight, portable and disposable medium just like I use and enjoy plastic beer cups. But I have much greater trust in16 bit uncompressed audio because I know what is in there and what is not.
Topic: 24bit or 16bit -> mp3? Which is better? (Read 33514 times)