Editorial - Audiophile High Resolution Recordings

Are they always what they seem?

There has been an uncomfortable whiff of what smells suspiciously at times like snake oil in the air at Pristine Classical this week - when what started off as an innocent analysis of a couple of faulty FLACs (downloaded from another website) developed unexpectedly into a wider investigation and consideration of what we get when we download what are (apparently) high resolution music files.

Spectrograms  Visual analysis of audio

First though I'd like to introduce you to the idea of spectrograms - graphic representations of sound where time runs along the x (horizontal) axis and frequency runs along the y (vertical) axis. Here the crucial third dimension of amplitude is represented by a spectrum of colour, where the brighter the colour on screen, the louder the sound at that time and frequency.

An example of this can be seen here - this is a single channel of a mono recording - in this case a section of the RAI Furtwängler Ring, Das Rheingold, as issued on regular CD by Gebhardt - from the sample on our website:

 

As you can just about see if you look closely, the bulk of the action is taking place at or below 3000Hz (3kHz), with quieter audio present between 3kHz and 6kHz. A dark stripe around 6.5kHz indicates a steep filtering at and around this frequency, and above it, very little music soon disappears into random noise, which at these higher frequencies we usually call hiss.

 

That's a standard CD - for higher resolutions the diagrams are similar, though the y-axis changes. Remember that the maximum musical frequency is half the sampling rate frequency - thus a 96kHz recording will go up to 48000Hz in audio, and a "192/24" recording tops out musically at 96kHz. (The 24 here refers to the digital bit depth, which is not under consideration in this article and denotes the amplitude range of the file.) 

 

 

Analysis 1  High resolution historic recordings 

 

The first high resolution recordings I've looked at were transfers of commercial open-reel tapes from the 1950s from a website popular among many collectors, and which claims very high quality - lossless downloads are available at 96kHz and 192kHz, for example. What does this mean? Well each doubling of the sample rate doubles the frequency range of a recording, and each doubling of frequency adds an extra octave of higher-pitched sound to what's otherwise available.  

 

However, it's pertinent to point out that we're already way off the top end of a piano keyboard here - C8, the top key on a concert grand, comes in at an apparently lowly 4.186kHz. It will however have several harmonics which can extend to an extent into these upper reaches of musical frequency.

 

This is where it gets contentious. Medical and experimental evidence suggests very few people have any ability to hear anything above 20kHz - one reason why CDs stop a little above here - and the vast majority of us, especially as we get older, have upper thresholds much, much lower than this. So is there any point in all of this ultra high frequency recording? And what might we be hearing, or missing, at frequencies traditionally reserved for bats? Is there anything meaningful "up there"? 

 

In the case of the aforementioned 1950 tapes, "not a lot" would appear to be the answer. I downloaded several free 96kHz (with an upper audio limit of 48kHz) examples from the website - you might expect these to be ideal samples - and found some highly unusual things going on:

 

 

(Apologies at this point for having to squeeze these pictures into a small space!) 

 

What you see above is the bulk of the music - from Sibelius's Karelia Suite - reaching about 15-16kHz, with a peak where percussion (and the high frequency distortion noise associated with it) reaches around 24kHz.  

 

You also see a black top half above the purple tape hiss - this recording was almost certainly transferred at 48kHz and later doubled up to 96kHz without any additional information being included - pay for it (and, if it exists, its 192kHz incarnation) and to put it bluntly, you've been duped. There's literally nothing there - the record producer has simply changed the file format, making it much bigger by adding a load of zeros to represent the absolute silence present in this file at the audio frequencies between 24kHz and 48kHz.

 

For the record, the first track I ever bought at 96kHz, from a different (and well known) music downloads site, and of a modern digital recording, showed exactly the same trick being used to increase the price without giving any extra "value".  

 

(I'm not naming names in this article because I really don't want to get into a naming and shaming exercise, nor any legal wrangles. What's here is here through pot luck in purchasing tracks or, in the case of a site which offers free samples, an analysis of those samples.)

 

Not all the samples from the 1950s tapes looked like this - here's a better example:

 

 

Again, though, there really isn't anything up there - above CD frequency limits - except hiss, at lower levels in upper frequencies than the tape hiss. Except in two places, and curiously only on the right channel (the lower of the two in this stereo file). Let's zoom in and take a closer look at one of them:

 

 

This is the second of two instances where there's clearly something going on - but alas it isn't anything you'd what to hear. This is probably clipping, where the output level has overloaded the analogue-to-digital converter, producing a click (or clicks) - essentially a very short, sharp burst of noise at equal levels across all frequencies. There's absolutely nothing up there which originated from the original 1950s tape except tape hiss, even when the recording is properly made (well, almost), and even this hiss starts to drop away above about 20kHz. 

 

 

 

Analysis 2  High resolution modern recordings @ 192kbps 

 

Moving on 50 years or so, what about modern, digital recordings? Surely these should make all the effort and expense worthwhile?  

 

The crème de la crème in generally available sampling rates is currently 192kHz (although some have advocated using 384kHz as a sampling rate), offering yet another octave, and a musical frequency response up to 96kHz. But is there anything actually there, whether or not you think you can hear it, and if so, what?

 

 

This is one channel of an orchestral recording - Fauré - recorded at 192kHz, with the audio frequency axis going up in 10kHz annotations up to a maximum possible of 96kHz. The orchestra itself peters out at about 30kHz (during its highest volume peaks), which is half an octave of very high frequency harmonics (or overtones if you prefer) above what can be accommodated by a CD - and which is also comfortably within the 48kHz audio frequency range of a 96kHz-sampled recording.  

 

Those horizontal lines you can also see are interference picked up during recording, possibly from fluorescent lights, computer monitors or other electronic equipment. Other than these all we have in terms of extra "value" for our 192k file is empty data and background noise.

 

 

From the same company, a 192kHz download of Winterreisse shows a similar picture: on the left hand side you can just about make out the relatively tiny frequency range of the piano; the singer comes in about halfway through this image, and it's his sibilance - the T's and the S's - which are the only component which reach up into the 25-30kHz range. Above this, yet again, there is nothing except some curious electrical interference at about 88kHz. (I found this an unusually and unpleasantly sibilant recording, by the way.) 

 

It strikes me, looking at these spectra, that if you're paying (through the nose) for 192kHz recordings you're probably buying a lot of empty space...

 

 

 

 

Analysis 3  High resolution modern recordings @ 96kbps 

 

So what about 96kHz downloads? Well, having found no real reason, in the realm of acoustic voices and instruments at least, to go all the way up to 192kHz, let's take a look at this level - and remember the top of these images now represents audio at 48kHz:

 

 

This purports to be Beethoven's 4th Piano Concerto (that much is correct), offered from a different label and downloaded from another website, and it claims to be recorded at 96kHz. Except that, if you look closely, once again it's not quite all that - it was clearly recorded at the lower rate of 88.2kHz and later "up-sampled" to 96kHz, hence the black stripe at the top. Nothing was ever captured in those highest frequencies, and thus there's an upper audio frequency limit in this recording of about 44kHz, rather than the 48kHz implied by the 96kHz download I thought I'd bought.

 

Mind you, I don't personally think I've lost anything worthwhile as a result. There are some impressive orchestral climaxes - as usual it's the brass and percussion which can just about generate sounds at these heights during their loudest moments. But take a look instead at the right hand side, where quieter strings and solo piano would barely trouble a late 1940s 78rpm disc, at least in terms of frequency extension.  

 

Yet again we also see a couple of horizontal lines, where electronic interference is present - though at too high a pitch to be heard, certainly by me (and I can just about discern very loud sound being switched on and off at around 20kHz in auditory tests). I might add that this kind of very high frequency "whistle" is present on many, many thousands of recordings across all genres, can be found on both CD and LP. It's usually at frequencies most of us cannot hear, or can barely hear when blasted at very high volumes, in isolation. But I've yet to hear a critic - or golden-eared audiophile - complain about it.

 

 

 

 

Analysis 4  CD resolution modern recordings @ 44.1 kHz  

 

One last FLAC download image, this time at regular, CD style, 44.1kHz:

 

 

Here, the very top end of this modern, digital recording the audio content has been rolled off above about 20.5kHz (filtering here avoids a nasty kind of digital interference and is quite common close to the upper reaches of sampling rates), and you can see that most of the action happens well beneath this. You'll also see a nice straight "whistle" line again, this time a little below 16kHz - probably picked up from a TV monitor somewhere near the recording equipment. To be precise, it's at exactly 15625Hz, the horizontal scanning frequency of a European PAL TV. The recording in question has been around on CD for quite a while, and was very well received by the critics - though again I've yet to see mention of this high pitched tone.    

 

 

 

 

Analysis 5  VBR (~200kbps) 44.1kHz MP3 

 

Finally, while we're at it, let's very quickly glance at an MP3, a modern recording from yet another very popular website, to see what we've got when we start losing data deliberately to reduce file size:

 

 

Looks very similar, doesn't it? This is solo piano, showing again that actually pianos don't really trouble the highest reaches of CD range. Look very closely indeed and you might just be able see some very dark - completely black, in fact - patches at and around the top of the image. This is the data compression in action - it's found nothing musical there, just very low level background noise, so it's been stripped out of the file. The compression has concentrated on retaining the obvious musical content alone, thus allowing the MP3 to be considerably smaller than the original file. Typically it's at this upper frequency level that MP3s discard data, which is one reason why they struggle with highly percussive material. The harder a file is compressed, the more dark areas you see - inevitably the more you lower the bit-rate the more this starts to impact on material you would like to hear.  

 

Very heavily compressed MP3s start to sound unpleasantly "squelchy", especially below 128kbps, whereas lightly compressed MP3s - such as our own 320kbps downloads - can be very open and compare very favourably to the source recording. 

 

 

 

So there it is - a little stroll through the joys and potential pitfalls of high resolution recordings. These tracks were all chosen at random from a variety of websites (none of them our own) and may or may not be indicative of a wider picture, but it seems that each and every one had an interesting story to tell the viewer - and listener...  

Andrew Rose, April 22th, 2011