When HD is a bit pointless

joerand, I am not sure really.. HDCD is a way of squeezing more of the original hidefefinition mix using Redbook bandwidth ie 44.1/16. It does this by non linear signal processing.

Therefore to create a HDCD master there needs to be a prior high definition master such as 48/24 or greater or an analogue source such as from a high speed tape transfer.

as far as I aware there are only two processors / AD converters that can create HDCD masters.. These being the Pacific Microsonics Model One and Model Two.. both getting long in the tooth now and out of production. The following HDCD mastering guide makes interesting reading on how to process the audio so as to be suitable for HDCD processing:

http://www.goodwinshighend.com...mixing_recording.htm

As far as SACD, well that uses DSD and not PCM of course ( exc any Redbook copy)... So it's not so much a format matter, but an encoding matter  .. So the whole recording/mixing/mastering/playback chain has to be DSD capable and therefore rather specialist and niche. I think it's fair to say because of this, one can assume attention to detail and fidelity is an important factor if one decides to produce media using DSD.

The message is trust your ears. Some of my HDCDs sound differently good. The majority sounded better back in my bare CDX2 days but pretty awful now. It's source material, system and listener dependent, same as always.

And regardless of bit depth and sampling rate you can't polish a turd. HiRes downloads are a crap shoot because if the file arrives intact you can't go after a refund if it sounds lousy, whereas Amazon and others will refund on a CD, DVD, SACD if you say the sound quality is unacceptable. 

Measurement and numbers don't tell the full story either. Last week's release of Transformer has a low DR and suffers for it. Last week's release of Berlin is also pumped but sounds rather splendid to my ears. If I knew in advance what Transformer would sound like I would have passed. Unfortunately there's no way a can commit less than 100% if it's a download.

I wonder if any unscrupulous organisation has yet taken an MP3 file and re-packaged that as HiDef?  I do think some are that bad.

There was reference to a certain Russian site here a few weeks ago that may well be converting mp3 to flac and selling it as flac.

@Aleg

AFAIK, I have only one CD mastered by Yoshida, a XRCD, The Joker by Steve Miller. It has some excellent attributes; depth, clarity, resolution and a black background. Probably a very true, flat-response. On the other hand these combine for a rather clinical sound that lacks warmth to my taste. My listening is primarily classic rock and I believe Yoshida is lauded more for his classical and jazz mastering with few credits in rock.

@Simon,

I've mentioned it here before, but the most recent HDCD production I've come across is Neil Young's Psychedelic Pill (2012). Unfortunately the format is all but dead as you note. I find the sound consistent across my catalog, seek out older HDCDs and buy with confidence. They can be found for cheap and I'd imagine if properly ripped would stream well.

@Harry,

My 60 or so HDCDs sound consistently good across the spectrum, but I'm listening on a 5x and something more resolving might prove otherwise. Still, they only got better going from a Nait XS to SN2. I agree you can't polish a turd, but the artists that took the cost to master in HDCD seemed to pay heed to the original recordings/production. On the other hand, you certainly can tarnish recording quality with compression and loudness (Pink Floyd's 2010 CD remasters and Zeppelin's Mothership come immediately to mind).

Part of my point in replying here is that it seems there might, selectively, be better routes to assuring a quality stream than downloading; buy a quality CD for a relative pittance and rip it to your standards. But that obviates the rationale of downloading doesn't it? Vinyl from any decade can be a crap shoot as well. As noted here, it comes down to the initial recording and mastering quality. That's a large reason I've learned to look at mastering credits when I contemplate purchases of physical media.

Interesting feature in this months Hi-Fi World about a PC/Mac analyser for verifying bit depth & data range.  Amongst many things it shows if the original data is genuine 24bit or 16bit upsampled & that the data extends out to 48kHz & does not stop at the 21kHz cut off filter of CD originated files.  

Would be nice if we had a www facility that we could use to read or test this info before purchase to avoid the fake "high res" downloads & hasten the demise of these scammers. 

Mike, really good idea, unfortunately I don't think it'd be easy to implement in Windows (I don't know for OSX).

For windows, to intercept stream data during playback of a file from a browser, I believe the component may have to be implemented as a KS Filter.

I think the browsers will call the User Mode WDM Audio Components, and these in turn call the Kernel Mode KS system directly.  No filter driver pins are exposed by the User Mode WDM Audio Components.

There are exposed pins in the KS Filter graph system, unfortunately KS Filters are Kernel Mode minidriver code, and that's not easy to implement with sufficient code quality to ensure system stability (errors can cause BSODs!).

The only other way would be to write a lightweight KS Filter that uses a gate transition in a private interface to reflect the data stream back to User Mode code for analysis, but that's just weird.

Wat,

Correct me if I'm wrong (and I may well be), but I thought Audirvana only played music from local resources not streamed from HTML web pages on the net?  This would need you to buy the file before testing it.

But yes your points about 4bits of dither (i.e. noise) and vinyl transferred to CD are absolutely correct.

Hmm, all sounds a trifle simplistic... a signal can be .. And is often zero sampled oversampled.. Therefore by definition there is no frequency information above the sample rate other than noise.. But the recording dither is averaged over the new oversampled signal so the the noise energy in the original signal band pass is reduced. Now true most over sampling DACs appear do this on reconstruction, but if you didn't use a zero sample oversampling algorithm such as an interpolated algorithm, , the hidef signal with zero sample oversampling with frequencies only upto the original sample rate / 2 will most likely sound better than the original non oversampled hidef signal.. This would  hold true for mixing and mastering as well.. Where the original digital master may be zero sampled oversampled prior to it being  mixed and processed.

Therefore checking frequency content is too simplistic.. You need to look at noise distribution as well and compare to the original non oversampled signal prior to mixing and mastering.

True better labelling would be preferable on hidef.. perhaps explaining the original recording resolution, and then mixing/mastering resolution... Kind of like the CD's DDD but for sample and bit length for recording, mixing, and mastering.

Simon

Simon, true, at least to the limit of my mathematical understanding.  However there are two domains to discuss:  Oversampling in the time domain, and bit extension in the amplitude domain.

Oversampling in the time domain could be done by simple interpolation and that could be relatively easily detected from the frequency distribution.  However, with well implemented dither, it would, as you say, be quite difficult to detect after the event as these data are correlated to the signal.

In the amplitude domain, putting a simple noise signal in the LSB of a 16 bit recording and claiming it as 24bit should be detectable as signal in the LSB will be uncorrelated to the higher 16bits.  Again if a true dither signal (correlated modulated nose) is used, then detection will be somewhat more difficult.

In both cases where a true dither is used, then, probably, an heuristic method would be needed to detect it.

Huge, but what are you trying to measure? Surely the key point is to determine whether the hidef signal within the original signa's band pass is worse or no better than the the original signal. Surely the only way to determine that is to compare to the original signal and / or listen for any betterment. Also is it not worth remembering  dither is a valid way of reducing noticeable quantisation noise in a digital signal to make it sound more 'natural' or musical.

Simon

Simon, it's not about faithfulness to the original data, or even about quality.

It's about whether the product is, in reality, as described.  That is, is it 24bit data rather than 16bit data repackaged as 24bit (with or without manipulation).  Similarly are the actual data 88.2khz+ rather than 44.1khz upsampled.  This is similar to the situation with digital cameras where 'software enhanced resolution' is no longer viewed as a valid definition of a cameras resolution.

16bit extended to 24bit, or 44.1 upsampled to higher rate, may be a contravention of the Trades Descriptions Act, or possibly the Sale of Goods act.  Test cases would be required in both cases.

Huge, like I said we need better labelling.. my point is there is no automatic 'technical' signature to a hidef file and over sampling can be a valid step in the production process.. Looking for ultra sonic frequencies to determine whether 'hidef' or not is flawed (in my opinion).. And I suspect will lead to mis leading results and I suspect false negatives..

What is required is an audit of the particular production process, and if fraudulent, then yes the necessary actions should follow.

Simon

Simon & Huge,  agreed determining high def is a bit flawed,  however IME there is a very definite signature with the distinct 21kHz brick wall filter of previously CD formated 16/44. The rest is somewhat open to interpretation - software diagnostics & interpretation skills is a whole other game - its not to hard to see upsampled 16bit, 16bit leaves a noise trace that 24bit does not,   & having a data trace to 48kHz is more indicative of reasonable quality high def than not.  

Going back to my previous post,  there are a number of dodgy downloads from well respected www sites such as Qobuz (for e.g.) that are clearly limited to 21kHz (CDA) & even 16kHz (MP3) despite their promises of real high def mastering.   What I am asking for is a www facility where downloads can be verified BEFORE purchase.

Hence my scheme for applying an heuristic determination to data in a playback audio filter graph from a browser.

From the heuristic you'll get a probability that it's derived from a hi-def master

I think that in some cases it can definitely be shown to be true (result = Accept)

In cases where the up-scaling and up-sampling are done crudely it can definitely be shown to be false (result = Reject)

In other cases however, it can be difficult to tell due to the nature of the recoded signal.

This could arise because:

The up-scaling and up-sampling are done properly and there are correlated components in both domains - this would lead to false acceptances.

The master contains a high noise level (e.g. Analogue primary masters or George J's contention that microphones should be placed many tens of meters from the performers) - this would lead to false rejections.

The music simply doesn't contain frequencies above 20Khz (e.g. some electronic music) - this would lead to false rejections.

In such cases the heuristic will give a probability between 0 and 1 as there is uncertainty.  It would then still be up to the consumer to decide how much they trusted the source (caveat emptor).  If a source gives a lot of indeterminate results, that would be a good indication of sophisticated cheating!

Thinking about it, the best approach may be to use a pre-trained naïve Bayesian filter.

Looking around the various www download stores I've found that High Res Audio (Germany & Japan) have some interesting pages on verifying download quality  (to support their quality promise)

The German site has some useful advice --- 

https://www.highresaudio.com/texte.php?ca_id=548

The Japanese site has more or less the same, albeit a different format ---  

https://www.highresaudio.jp/in...&page=en.spectre

However the Japanese people publish a spectrum analysis of the album you are about to buy,  you need to read the link above to understand it,  but this is what you can get at the touch of an icon button - before you buy

While it's a clear story it's in real life still confusing. For example take the actual example of an album which is coming out now. The new album of Cassandra Wilson is coming out on Qobuz in 24/44,1 on high res audio Germany it's offered on 24/96. So how would I now check that the 24/96 one is better and not just an ups called version. The price difference is not huge - but there is a curiosity to understand when you are being mistreated and when not.

anybody some light on this perhaps - using the actual example ?