Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: High Res Music. Where does the data come from? (Read 4813 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

High Res Music. Where does the data come from?

I've read all the arguments against high res music, and I agree with all of them.  I've done an abx test I posted here earlier showing that I couldn't tell the difference between a 16/44 FLAC and a 24/96 FLAC.

A lot of the super high res stuff in the 24/192 category are old recording done in the analog age.  If 16/44 pretty much covers the range human hearing, where does all the extra data come from on these recordings?  I know it's data I can't hear.  I'm just not sure how it gets there.  Even though we can't hear it, is recording equipment still sensitive enough to collect it?  And even if it is, is old analog studio tape of sufficient resolution to capture this range?

I also tend to wonder what a human would hear with these files if they could.  Would the extra range of sound be pleasing to the ear, or grate on your nerves and ruin the song>

High Res Music. Where does the data come from?

Reply #1
The data comes from the a/d converter. The "extra" stuff is noise and extended (for a lack of a better word to indicate a comparison) frequency response. The latter may or may not exist on the analog source tape.  It's highly unlikely the former will with any measurable degree of precision.

High Res Music. Where does the data come from?

Reply #2
It is possible for at least some extra information to be recorded vs a 16/44 FLAC or CD, possible for older recordings to be remastered (with varying results depending on the available sources), possible even for some equipment to faithfully reproduce that extra information (although you probably don't possess that equipment).  Whether any of those things happens or not can be difficult to find out.

Assuming for a moment that a hi-rez digital file can be made from a suitable (re-)master, quite what you might hear is hotly debated.  The evidence is pretty thin that you will be able hear any difference between properly made 16/44 and 24/96 or higher encodings - at least without further non-auditary cues.  Some of the evidence suggests that any audible difference will be to the detriment of the audio.  If you want to dig further, first read the many many posts about this already on these forums, then consider that the two components of a "hi rez" encoding are almost entirely unrelated and should be considered separately.  The first number is a bit depth and changing it largely affects the noise floor, with 16 bits having a high enough noise level to be theoretically audible although in practice it never should be.  The second number is the sampling rate and if affects the ability to record higher (inaudible) frequencies, with the side effect that increased noise propagates down to lower (audible) frequencies.

 

High Res Music. Where does the data come from?

Reply #3
Would you mind elaborating upon/substantiating this side effect?

Specifically, it seems you're stating that it will rather than just that it can.  Mind you, I'm interested in what is audible, not what is theoretical, though I'm also interested in what is actually measurable.

High Res Music. Where does the data come from?

Reply #4
We're not necessarily talking about upsampling, but if you simply upsample from 44.1kHz to 96kHz, you'll have additional interpolated data-points and data points at different time-points.  There's no more meaningful information, but there are more bytes. 

And in the end, your digital-to-analog converter is going to create a continuous analog waveform with essentially an infinite number of points.   

If you convert from 16-bits to 24-bits (or more) without changing the sample rate, you simply have 8 additional zero-bits....  Something like writing 24.0, or 24.000000    The second number has more precision and it takes more bytes to store it, but there is no more information.

If you do both, and upsample from 44.1/16 to 24/96, the interpolated data will have the full 24-bits of precision...  You won't have 8 bits of zeros in every sample...    But, you'll have no real usable-meaningful  information.  And again, your DAC is going to put-out a continuous analog signal anyway.

Or, if you want to take an easier example, a 48kHz 16-bit file upsmpled to 96kHz has exactly twice as many bytes.  And if we upsample it to 24/96 we have 2.5 times as many bytes.  (That's assuming uncompressed files.) 


Quote
is recording equipment still sensitive enough to collect it? And even if it is, is old analog studio tape of sufficient resolution to capture this range?
As far as I know, there are no analog tape machines with the equivalent of 16-bits of dynamic range.    So, any low-level sounds around the -80 to -90dB will be masked (drowned-out) by noise.  So yeah...  The signals are recorded, but you can't hear them through the noise....  Imagine that you are out in your yard and you are speaking in a normal voice to your neighbor who is several feet away...  He can hear you just fine, and he could even record you.  But, then a construction crew in the street starts running diesel engines jack-hammering...  Now, the sound waves from your voice are still hitting his eardrums and going-into the microphone and being recorded...  But your voice is permanently lost in the noise. 

Quote
I also tend to wonder what a human would hear with these files if they could. Would the extra range of sound be pleasing to the ear, or grate on your nerves and ruin the song
Well...  On the one hand, there's no reason for it to be pleasing, or anything but noise.

And in the real world if you use dynamic compression to bring-up the quiet sounds, the noise often  becomes more noticeable and more objectionable, especially if you are starting with a "noisy analog" original. 

On the other hand if we could hear, it musicians & instrument makers (and engineers & producers) would attempt these quiet details sounds pleasing.

High Res Music. Where does the data come from?

Reply #5
where does all the extra data come from on these recordings?


Could be a fine mess of the following:

- recorded sounds. Here are some tape machine frequency response curves: http://www.endino.com/graphs/

- tape hiss. With 24 bit resolution you are going to capture a lot of that. And whatever is "just noise" is hard to compress, likely increasing filesize by way more than the 50 percent you would expect from going from 16 to 24.

- at worst: tape bias! Unless it is filtered away. I am not so sure if you will get that in a 96 kHz file though, I do not know if there was anything as low as sub-48 bias in use, apart from a 41 portable mentioned at http://richardhess.com/notes/2008/02/02/ta...as-frequencies/ - but definitely in 192, as there were a few 60 kHz signals.


Even though we can't hear it, is recording equipment still sensitive enough to collect it?  And even if it is, is old analog studio tape of sufficient resolution to capture this range?


I do not know about what was actually used in recording studios (the above graphs are cropped down to 20 kHz but you see they do largely not drop in response), but certainly there was technology available to record - and store on tape - signals way beyond the audible limit: http://richardhess.com/notes/formats/magne...nstrumentation/


I also tend to wonder what a human would hear with these files if they could.  Would the extra range of sound be pleasing to the ear, or grate on your nerves and ruin the song>


You can get an idea by slowing down, that was the only approach if you wanted to listen to bats :-)
Here you have some signals at 1/10th speed: http://www.zachpoff.com/diy-resources/exploring-ultrasound/

High Res Music. Where does the data come from?

Reply #6
The largest issue with all this 'high resolution' nonsense is because people see these inaccurate representations of digital audio that look like 'stair steps'. Some people even claim they can HEAR the 'graininess' of the 44,100 times per second the audio is being sampled! Meanwhile the DAC only puts out analog signals.  So 'high resolution' was created to address the 'fat' stairsteps and make them 'thinner' and 'more accurate to the original wave'. (also to re-sell music) See Montys video on digital audio from xiph.org for why this is way off base.

High Res Music. Where does the data come from?

Reply #7
For those of you with pro recording gear, how many bits remain unchanged when you short the input and record to 24 bits with the gain calibrated for line-level?

High Res Music. Where does the data come from?

Reply #8
I've read all the arguments against high res music, and I agree with all of them.  I've done an abx test I posted here earlier showing that I couldn't tell the difference between a 16/44 FLAC and a 24/96 FLAC.

A lot of the super high res stuff in the 24/192 category are old recording done in the analog age.  If 16/44 pretty much covers the range human hearing, where does all the extra data come from on these recordings?


Depends on the source, and the procedures used.

If the source is already digital, then the best practice would be to upsample it using a good resampler. In that case the recording is unchanged, just spread over more bits. There will be something in the new recording above the Nyquist frequency of the original recording, but it will be vanishing.

If the source is analog, then the likely practice would be to digitize it using a good ADC. In that case the recording should be  unchanged over the bandpass of the ADC, but it will be brick wall filtered at the Nyquist frequency of the sample rate chosen. Analog recordings do have some content extending up to infinite frequencies, but due to the technology used it is usually vanishing. Good practice would be to analyze the transcription and remove any obvious noises due to technical flaws that are large enough to fully mask any possible content at those frequencies.

Quote
I know it's data I can't hear.  I'm just not sure how it gets there.  Even though we can't hear it, is recording equipment still sensitive enough to collect it?


These days you can buy a 24/192 ADC based audio interface with dynamic range > 110 dB for maybe $150-300. There simply was never any analog media or legacy digital recordings whose dynamic range or useful bandpass comes close to that. Therefore even recording equipment readily available to a serious hobbyst can collect anything useful and lot of things that are not useful off of any conceivable legacy or modern recording.

Quote
And even if it is, is old analog studio tape of sufficient resolution to capture this range?


Not a chance, even the latest greatest with all of the usual enhancements including things like Dolby S.

Quote
I also tend to wonder what a human would hear with these files if they could.


I suspect that a lot of commercial transcriptions of legacy recordings were well made, so you might just want to find some and just listen to them.

Quote
Would the extra range of sound be pleasing to the ear, or grate on your nerves and ruin the song>


That is up to the skills and taste of the production staff performing the work. Transcribing a SOTA analog recording made in the 1970s or later using even mediocre modern technology is far from rocket science. The skill testing challenge is making recordings from say pre WW2 or even the early 50s sound acceptable to modern ears.

High Res Music. Where does the data come from?

Reply #9
For those of you with pro recording gear, how many bits remain unchanged when you short the input and record to 24 bits with the gain calibrated for line-level?


IME the low order bits of the commonly-used sigma-delta  converters operated under those conditions are pseudo-random noise and continuously vary accordingly.  IOW they are always changing, so none of them are unchanged.

The higher order bits above the noise floor will be steady zeroes, so they are indeed not changing. Sigma Delta converters usually have highly predictable performance because they basically operate in the digital domain. Its not uncommon for them to meet their specs once you figure out what the specs actually mean. Most of the analog buffer chips used with them don't add meaningful amounts of noise.

High Res Music. Where does the data come from?

Reply #10
I asked how many, though now I realize I should have asked about the shape of the noise.

I was trying to come at it from a different angle from the equivalent number of bits for analog tape.

High Res Music. Where does the data come from?

Reply #11
where does all the extra data come from on these recordings?


Could be a fine mess of the following:

- recorded sounds. Here are some tape machine frequency response curves: http://www.endino.com/graphs/

- tape hiss. With 24 bit resolution you are going to capture a lot of that. And whatever is "just noise" is hard to compress, likely increasing filesize by way more than the 50 percent you would expect from going from 16 to 24.

- at worst: tape bias! Unless it is filtered away. I am not so sure if you will get that in a 96 kHz file though, I do not know if there was anything as low as sub-48 bias in use, apart from a 41 portable mentioned at http://richardhess.com/notes/2008/02/02/ta...as-frequencies/ - but definitely in 192, as there were a few 60 kHz signals.


Even though we can't hear it, is recording equipment still sensitive enough to collect it?  And even if it is, is old analog studio tape of sufficient resolution to capture this range?


I do not know about what was actually used in recording studios (the above graphs are cropped down to 20 kHz but you see they do largely not drop in response), but certainly there was technology available to record - and store on tape - signals way beyond the audible limit: http://richardhess.com/notes/formats/magne...nstrumentation/


Usually the largest irreducable source of HF losses is probably due to losses related to the gap in the reproduce head as shown here:



This is taken from http://www9.dw-world.de/rtc/infotheque/mag.../magrec_05.html which lists most of the other sources of HF losses with pictures.

More details on the math here:

http://home.comcast.net/~mrltapes/mcknight...th-response.pdf

More math models of other sources of loss here:

http://www.aes.org/aeshc/pdf/mcknight_some...conceptions.pdf

The curved lines on the right the illustration above are truely asymptotic to zero. They repeat at harmonic intervals as the frequency goes up but the first one is so pervasive that nobody tries to go above it.

In real world tape pro machines, the first null might be between 22 and 30 KHz @ 15 ips.

High Res Music. Where does the data come from?

Reply #12
I asked how many, though now I realize I should have asked about the shape of the noise.


The noise floor of commercial ADC chips for audio is usually spectrally flat or white.

I've never seen a device that performed otherwise, and I've never seen one that offered any other choice.

A consequence of this is that if you wanted to use noise shaping during recording you would have to use an ADC with vast overkill performance and do the noise shaping during downsampling.

The counterpoint is that most pro audio gear used for recording has a noise floor that is spectrally flat and in the general vicinity of 96 dB down.

I know that there have been a few hardware products including one by Meridian (the 518)  during the 1990s that offered noise shaping options, but I don't know how it was implemented.

More info here:

http://www.stereophile.com/content/meridia...audio-processor

Interesting quote:

"The 518 is a successor to Meridian's 618 Mastering Processor, which has gained no small reputation in professional audio circles for the way it reduces 20-bit audio data to the 16 mandated by the CD standard without sacrificing sound quality."

I'd like to query JA and JRS about how much they stand behind this product and its technology today, as it would appear to eliminate the need for > 16 bit sampling if their claims at the time were true.

The noise shaping options in CEP appear to me to possibly have been patterned after it.

High Res Music. Where does the data come from?

Reply #13
Ok then. How many LSBs are changing in a typical pro recorder?  I'm (still) looking for a numerical answer. A smallish range of numbers will be ok if you're too uncomfortable taking a stab at a single number, though I would also like to see a stab the median number.

Again, input shorted, output set for 24-bit LPCM, gain calibrated for line level.

Anyone else care to speculate, if not provide real data?

High Res Music. Where does the data come from?

Reply #14
Ok then. How many LSBs are changing in a typical pro recorder?  I'm (still) looking for a numerical answer. A smallish range of numbers will be ok if you're too uncomfortable making a stab at a single number, though I would also like to see the median as well.

Again, input shorted, output set for 24-bit LPCM, gain calibrated for line level.


Lower end pro recorders and audio interfaces have approx. 100 dB SNR in 24 bit mode.  Call it 17 bits.  That means the other 7 low order bits are noise.

Higher end stuff  comes in around 115-120 dB IOW approximately 19-20 bits. That means the other 4-5 low order bits are noise.

High Res Music. Where does the data come from?

Reply #15
So to the original poster:
You can see some of the data may have* nothing to do with the original source; the "data" (read: noise) can solely be generated within the analog-to-digital converter itself.

Thanks, Arny!

(*) I'm being overly-generous with my choice to say "may have."

High Res Music. Where does the data come from?

Reply #16
So to the original poster:
You can see some of the data may have* nothing to do with the original source; the "data" (read: noise) can solely be generated within the analog-to-digital converter itself.

Thanks, Arny!

(*) I'm being overly-generous with my choice to say "may have."


Right, because we are now at the truism/nit picking level. No real world process and even many theoretical processes are totally free of noise and/or distortion.

The very process of transcribing something implies processing and a format change and that strongly implies the addition of at least trivial amounts of noise and distortion.

To show how nit-picky this can be, a very high quality transcription of a LP or analog tape implies the addition of what most sane people would call  trivial amounts of noise and distortion, but that can easily be less than the noise and distortion added by simply playing it one more time.

High Res Music. Where does the data come from?

Reply #17
For those of you with pro recording gear, how many bits remain unchanged when you short the input and record to 24 bits with the gain calibrated for line-level?
Depends on the sampling rate. HF noise from the modulator increases above a certain frequency. My Lynx Aurora at 24/96 (line out to line in) has the 6 lower bits active, the higher ones remain unchanged. NB: this is DAC noise + ADC noise. The ADC dynamic range is specified 117 dB(A) by the manufacturer.
What would you like to do with this information ? Active bits are not necessarily related to the dynamic range in a specific frequency band (just think of DSD).

High Res Music. Where does the data come from?

Reply #18
A few thoughts:

Noise is generally specified in terms of total energy over the audible bandwidth, typically 20Hz - 20kHz. Quantization noise will have a rising 6dB/octave (white noise) profile. This tells us that the noise level in smaller bands of interest is somewhat lower, but will tend to increase with the frequency of the band in question. So, even in a 16 bit system with a SNR of ~96dB, signals below -96dB can exist, especially at lower frequencies. Also, given a properly dithered signal, signals below the noise floor are represented. And, in terms of psychoacoustics, it is possible to perceive signal which is below the noise floor.

Thoughts?
Harold Kilianski

High Res Music. Where does the data come from?

Reply #19
Noise is generally specified in terms of total energy over the audible bandwidth, typically 20Hz - 20kHz. Quantization noise will have a rising 6dB/octave (white noise) profile. This tells us that the noise level in smaller bands of interest is somewhat lower, but will tend to increase with the frequency of the band in question. So, even in a 16 bit system with a SNR of ~96dB, signals below -96dB can exist, especially at lower frequencies.


This is true.  -96dB is the spectrum integrated time domain noise floor.  The actual noise power spectral density (that is, noise per Hz) is far lower, meaning that relatively pure tones far below -100dB are still well resolved.

Also, given a properly dithered signal, signals below the noise floor are represented. And, in terms of psychoacoustics, it is possible to perceive signal which is below the noise floor.


Psychoacoustics aren't even needed here.  Take one second of white noise at 44.1khz, and at it to a sin wave with one tenth the RMS power and take an FFT:



Not bad for a signal at -10dB SNR.

Edit:  Sorry, that initial post sounded like I was saying you don't need to dither.  Not what I meant!

High Res Music. Where does the data come from?

Reply #20
I've read all the arguments against high res music, and I agree with all of them.  I've done an abx test I posted here earlier showing that I couldn't tell the difference between a 16/44 FLAC and a 24/96 FLAC.

A lot of the super high res stuff in the 24/192 category are old recording done in the analog age.  If 16/44 pretty much covers the range human hearing, where does all the extra data come from on these recordings?  I know it's data I can't hear.  I'm just not sure how it gets there.  Even though we can't hear it, is recording equipment still sensitive enough to collect it?  And even if it is, is old analog studio tape of sufficient resolution to capture this range?

I also tend to wonder what a human would hear with these files if they could.  Would the extra range of sound be pleasing to the ear, or grate on your nerves and ruin the song>



Tape bias signal is one thing that would require 'hi rez' ADC to capture.  That inaudible signal has  been used for digital 'flutter correction' by Plangent.