I'm curious about the claims asserted by Head-Fi member Ian in this Head-Fi thread:

http://www.head-fi.org/forums/showthread.p...989#post3009989

I'm looking specifically at posts 7 and 15 of the thread.

He asserts that MP3s and other lossy files are "best played back at 24 bit." In support of that assertion, he claims that (1) when files are compressed to a lossy format, the bit depth is reduced, and (2) that the files are therefore better off being played back at 24 bit.

The claim (1) seems incorrect to me. Does lossy compression affect bit depth? Even assuming that his assertion in claim (1) is correct, his conclusion in number (2) doesn't seem to me to follow logically from (1).

Comments?

He doesn't say anywhere that the bit-depth is reduced. In fact, he seems to understand pretty well that bit-depth no longer has any meaning once something is compressed to mp3.

While it may be true that decoding to 24 bits give a more precise representation of the encoding than 16 bits, I have a hard time believing that this is going to make your mp3s sound better.

There is the obvious example of a very low level sound that is encoded from a 24 bit original (meaning it can't be from a CD). The MP3 will not have a problem with the dynamic range, but playing back at 16 bits you could end up with silence, which wouldn't happen with 24 bits.

In normal usage this is not a problem. Properly dithered 16 bit playback will give 115dB of audible dynamic range, enough for the human hearing system. 24 bits will not produce an audible benefit (if you disagree, please try the MAD challenge), but *if* your playback chain is actually capable of >16 bits playback, then it would still be preferred to use 24 bits for the simple reason that proper dithering requires some extra work from the CPU.

As for "MP3's have no bit depth": MP3's store the sound as amplitude over frequency, whereas a normal WAV has it as amplitude over time. The two are not directly comparable. In reality, most MP3's have a variable bit depth which often goes as low as 1 or 2 bits of ACCURACY, but with a much larger RANGE (in AAC the range is 64 bits, I think it's the same for MP3). It's because of this large RANGE that there is a _theorethical_ benefit to 24 bits playback.

Thank you for pointing that out. I did misread his post in that respect and your explanation helps me understand his point.




The point that piqued my interest in the Head-Fi thread is that a CD burned from an MP3 that was encoded from a CD (i.e., 44.1hKz, 16 bit original) will have a lower quality than the MP3 itself. I was under the impression that the CD would be identical in quality to the MP3 from which is created.

When the MP3 is decoded to PCM it can be done as 24 bit integer, 32 bit float, etc., and with the proper hardware and software this added bit depth can be maintained through reproduction of the sound. If instead there is a conversion to 16 bit PCM to go to a CD then even with proper dithering there is theoretically a reduction in quality. This reduction is quality is most likely not audible.

Except that if it came from16 bitsin the first place thereisno "this added bit depth" to go around.

You will loose something (depending on your dithering or noise shaping method of choice) on the conversion back to 16 bit. LIke others, I doubt this is audible.

I'm concerned about mp3 bitdepth.

I made a post about this in Apr 29 2004:
mp3 mastering/encoding, 24 or 16 dithered

The question was quite different though: Being able to have a master at a higher bitdepth than the traditional 16 bits, could I get some benefits encoding a mp3 from 24 or 32?

Actually, my thought/feeling/experience was that 32 bit float is probably the best alternative, avoiding the accumulation of two dithering (dithering at 16 at the mastering and dithering at the playback of the mp3) while the dithering can be made at the bitdepth of the souncard. Also the encoder don't have to encode the dithering of 16 bits which in this case could be unnecessary. Actually with lame aps, a small size's difference was noticed.

Now an interesting question might be: can we hear the difference between an mp3 encoded from a 16 bits dithered and an mp3 encoded from 32 bits float undithered.?
If yes, which one is closer from the original (at 32 bits float)?

An ABX might be harsh on this…
My impression (= no ABX ) with a lame at API was that the mp3 from the 16 bits was more aggressive and the mp3 from the 32 bits had more depth. So in terms of mastering, it could be decided according to the sound you want for your release on mp3…

I'm sure that a test can be rigged up such that an MP3 decoded and played back as 24-bit PCM data will sound noticeably better than as 16-bit PCM data. But you'd have to rig such a test if you want easily ABX-able results. By rigging, I mean you choose a specific kind of sample to test with. I would suggest a source consisting of a super soft, simple 16-bit sine wave with 1 "pixel" amplitude unit, and turn your amplifier super loud so that you can easily hear this sound (and you'll probably hear lots of crap as well, that's the point of the test, because the S/N ratio of the original file will be 1:1 under this condition).

In practice, I doubt there is much difference, like most other people here said. But theoretically the quality should be better. I'm not one of the people whose philosophy is "if you can't ABX the difference, there is no difference." And as long as there is a theoretical improvement in quality, why not play back as 24-bit PCM if you can, for free? That being said, I myself am too lazy to do that and doubt I would ever be able to tell the difference in a practical listening situation, so I just let my players do whatever they do by default which is probably 16-bit playback. The source CD was 16-bit PCM data to begin with anyway, I don't really care that much.

You can easily prove an objective measurable difference, and create samples which give a clear audible difference if the volume is boosted on playback.

This isn't to dispute the fact that it's probably inaudible under normal circumstances.


Often, 16-bit sources encoded to mp3 remain approximately "self dithering" at the 16-bit level, but this isn't always the case. Sometimes the encoder removes the low level noise, and sometimes it removes high frequency noise shaped dither. This can leave a signal with a noise floor below the 16th bit (i.e. not self dithering). If you don't dither when decoding, then you are effectively truncating a higher bitdepth source, which isn't advisable. If you do dither, then non noise shaped dither could increase the noise floor wrt to the original, and noise shaped dither could increase the peak amplitude, possibly leading to (extra) clipping.

So the best option really is 24-bit decoding. (And, of course, some mechanism of preventing clipping, e.g. ReplayGain clipping prevention). I assume that's what most of us are doing with 24-bit sound cards and foobar2k. There's no reason not to.


Not only does it not matter, but at 16-bits, against all theory and advice, I prefer to decode mp3s without dither because its very hard to make even an unrepresentative pathological sample where the truncation distortion is audible, whereas its slightly easier to make an unrepresentative pathological sample where the noise is audible. Even so, it really doesn't matter.

Cheers,
David.