I remember that when the Digital Compact Cassette was launched (circa 1993) there were claims that PASC had higher audio quality than PCM (16 bit @ 44.1 Khz). I thing i read that on "Stereo Review" magazine. I fact i remember that they were saying that the DCC was number one, CD was number two, and MiniDisc number three.

Was this true?

Is there any lossy audio codec (at any bit rate) that can beat PCM 16b @ 44.1Khz?

A quick google research for PASC+DCC revealed that PSC is essentialy MPEG-1 layer 1, a lossy encoder, so I strongly doubt it can be "better" then PCM. At most it could be "musically transparent", while with PCM you can encode whatever signal at your heart, with the inerent precision given by the bit depth and sampling rate chosen.

Sergio

The point is not "can we have a better quality?" but "can we hear a better quality?". I personnally do not want to disturb bats, dogs or whales with some *audiophile* high/low frequencies.

I have several DCC machines, actually. I love obscure, unknown formats. I even created one.

The reason that they claimed that it was better than 16/44 was because they used 18 bit resolution. Their logic was that the improvement of having 18-bit resolution would be greater than the loss caused by MPEG encoding. I'm not sure I would buy that (the final bitrate was about 384 kbps I think), but I do think that it's possible to have higher quality than uncompressed 16/44 using a lossy codec. See this post of mine:

http://www.hydrogenaudio.org/forums/index....ndpost&p=321039

Simply impossible. What is lossy audio compression but taking away digital information from a PCM stream?

I often like to compare to the still images realm to see if a theory works. Is there any lossy image compression format that actually produces better output than a 24-bit BMP file? Of course not. The analogy goes further: since the human eye can discern only about a million colours, 24 bits per pixel (16,777,217 colours) are perfectly sufficient, and those, erm, shall we call them "videophiles" who push for "high definition", 48-bit images, are simply deluding themselves that it's better. Of course 24-bit audio is more faithful to the analog original than 16-bit audio; but if that faithfulness lies in reproducing frequencies that are outside the range of human hearing, then it has as much value as those 48-bit images.

Hmmm ... I seem to have just made a longer elaboration of what damaki said, about the importance of hearing rather than having better quality.

This depends on what on is doing with the audio. 24 bit, or better 32 or 64 bit floating point, is very useful for improved quality while one is in the working-on stage with new music.

Also, a few musical instruments produce significant signal up to 30kHz or so. While it is generally accepted that people cannot hear those frequencies, there was a study, from Japan I think, that found two different kinds of increased brain wave activity in subjects during blind AB tests when listening to such music with the higher frequencies present (as compared to the same music lowpass filtered at 20kHz). The subjects were not aware that differences were being tested on them, but they reported liking the music better at some times than at others. These times turned out to be when the full range of high frequencies were present.

The published study I read was preliminary, that is, apparently the first such testing done. It was thus hardly conclusive, but quite interesting. I read this report off the web somewhere. Unfortunately I have been unable to find it again. If someone knows about it, I would appreciate a link.

Actually, it's not quite as simple as that; there is a reason why 48bit images are useful. Not for viewing, of course, as you rightfully point out, but for editing. If you're going to be doing major adjustments to the colours or brightness levels of an image you want it to be as high a bit depth as possible. If you've ever seen a photo that was very dark and seen it brightened up severely, you will notice that a 24 or 16 bit image results is a lot of banding (as the small numbers are now large, badly rounded numbers), whereas a 48bit image (or 16bit per channel, as they call it in digital photography) can still preserve reasonable accuraccy when scaled by such large amounts.

Likewise with audio. I, too, agree that most people who claim to can't actually hear the difference between 16 and 24 bit; however, I would still work at 24bit in a Digital Workstation when recording and mixing, as any gain or processing you apply will accentuate the rounding errors and the end product may only be as good as a 11 or 12 bit signal. Simply put, save all rounding to one step at the end.

Hence high resolution is useful (just not really for an end listener).

one other possible argument for 16bpc images (i wont say 48 bit because most likely we're working with subsampled chroma, which complicates things this way) is that 8-bit _linear_ images may not be sufficient depending on the response of the display used.

TVs produce drastically boosted darks compared to computer monitors. LCD displays do even more bizarre things to images.

viewing a properly calibrated TV in a dark room is a great way to spot artefacts...

though 8-bit logarithmic images are probably sufficient for that purpose.

another thing is not a matter of how many _colours_ we can distinguish, but more a dynamic range thing - our eyes are constantly adapting to the subject, and hence we can see detail both in brightly lit clouds on a sunny day and the shadows in a wooded area by simply moving our eyeballs a bit. 8 bits is absolutely nowhere near enough to represent that kind of dynamic range, even as a delivery format. in fact, most film and definitely CCDs can't do that (with the possible exception of ansell adams, but he's long gone).

one last thing - MPEG-2 isn't 8 bit per channel by nature (though it's usually decoded and even encoded as 8 bit), but will take 9 bits and is most often fed with a 10-bit source. so maybe the videophiles are getting what they want to a small extent. now the trick is getting a display that can come even close to 9 bits... plasmas are nowhere near that.

*PASC was not 18bits. Pasc itself is mpeg1 LayerI, so it is bit independant. The 18bits thing was probably related to the DAC/ADC, not the coder itself.

*MPEG coders quality can always be surpassed by using PCM if using proper resolution/sampling rate. However, mpeg coders can theorically be better than 16bits PCM, as they have an higher dynamic range. Of course this is only if you feed them with something higher than 16bits as input.

Yup, Carlos, I was aware of that. And I'd use 16 bits per channel myself if my camera supported it. Unfortunately, it's an old Ricoh RDC-7, and 8 bits per channel in TIFF mode is the best it can do.



That's the argument audiophiles make for 24-bit, 96000 Hz audio: that it offers dynamic range, and thus the "warmness" that is, allegedly, the characteristic of them vynil recordings.

Has anyone ABX'ed to check that so-oft-made claim?

But if the original source isn't 44.1kHz 16-bit PCM but something higher quality. I didn't see anywhere the requirement to lossy compress the same signal.

Perhaps they should consider turning on their heaters instead.

So if the source is let's say 24b/96Khz, a lossy codec could be perceived as having better quality than PCM 16/44.1?

Then it would be better than 44.1/16/2 PCM, but not by virtue of the compression scheme--only because of a better source. Which source itself, most probably, would be PCM, thus invalidating the whole argument.

I wonder: apart from PCM, what is there? Lossy compression formats are derivatives of PCM. PCM is as basic to the audio world as RGB pixel-mapping is to the visual one. In the visual world the only other game in town is vector graphics, and even here the instructions translate to RGB pixel-maps in the end; likewise, MIDI calls to samples on the sound card, but they too turn out to be PCM data. The only thing I can think of outside the PCM paradigm is FM synthesis, and I think we can all agree there's no way that can match, let alone surpass, the quality of PCM-mapped recorded samples.

PCM is nearly 70 years old, by the way.

My impression was that the original poster wanted to know if _any_ (realistic) lossy codec can be better than the specifically mentioned CD-quality PCM. Perhaps I misinterpreted the idea.

DSD?

Isnt that bascialy PCM with aggressive noise shaping?

Aha. Thanks. Looks esoteric, though.

I'm not the right person to answer that as I'm not an expert but I was under the impression it is totally different from PCM. Huge sampling rate and each bit meaning a rise or drop in the signal - or something like that(?).

DSD = PDM not PCM (unless you're using a very liberal definition of PCM...)
http://en.wikipedia.org/wiki/Pulse-density_modulation

To answer the original question, yes it is possible to make a lossy codec that could be better quality than 16/44.1, providing your original source was of high enough quality (like producing 14bit ADPCM from a 20bit PCM source for instance) . If bryant is correct and the final bitrate of this format was only 384kbit: not bloody likely

In the days that I was working for Philips Classics, they did quite some listening tests to improve the PASC system. More and more audio masters in those days had 18 or 20 bit resolution (44.1 kHz), especially Decca was keen on 18 or more bit recordings.
The claim that a DCC made from a 20-bit master might sound better than a 16 bit cd does seem justified to me.

I don't think it's very justified, because lossy psychoacoustic compression will in fact with throw away many of those bits. Also, using more than 16 bits hardly makes a difference under usual listening conditions.

Edit: well, thinking more about it, if we assume that mpeg layer 1 encoding doesn't throw away data basing on the absolute loudness of the signal, then it could have an advantage in faint mussical passages. But only if the decoder uses DACs of greater than 16 bits of resolution.

My DCC recorder had a 18bits DAC if I remember well.

But all this is mainly a theorical advantage. Real life might be different.