BenLaw
Well-known member
Yes, difference plot of their 16 bit v their 24 bit and their 16 bit v downsampled 16 bit would both be interesting.
shadders
Well-known member
fr0g said:CnoEvil said:
Excellent.
Ok, as a test, I have now bought that track from Linn (2.5 euros)
It was a 40 MB download.
I have downsampled it to 16 bits. The original is only 48 KHz anyway
I have compared the 16 bit version with the 24 bit studio master.
The difference plot...
http://i.imgur.com/s0cC0g1.jpg?1?6669
(full size image)
As you can see the difference is some sound at -76 dB and beginning at 17.5 KHz. Quite inaudible.
Conclusion, the 24 bit 48 KHz recording, and subsequently the 40 MB download requirement, is a waste.
There is absolutely no audible difference between the file at a lower resolution and the one they supply.
Hi,
Is it possible that the 24bit 48kHz download is just an resampled 16bit 48kHz file ?
Or that if you actually examine the 24bit values - the upper 8 bits are never used ?
Regards,
Shadders.
shadders said:
No, as you see, there "is" a difference..It's just that all the information is from 17.5 KHz and at -74dB or lower.
CnoEvil said:
Quite happy to do the test, but I'd rather not have to pay again. If you have the 16 bit version you could stick it on dropbox?
shadders
Well-known member
fr0g said:
Hi,
Ok- can you provide the spectral plots of both files ?
Thanks and regards,
Shadders.
shadders
Well-known member
fr0g said:
Hi,
In thinking about this - the difference is -90dB across the main audio band.
Assuming these are voltage measurements, this equates to 31uVolts difference across the audio band between the two files.
As such they are different across the audio band.
If there was no difference across the audio band, then there would be infiite negative dB difference.
Hmmm, then - these are samples you are processing - is that correct ?
Can you confirm what software you are using ?. Thanks.
The difference could be due to floating point representation in the software you are using. Just don't know until more information is provided.
Regards,
Shadders.
fr0g said:
I don't blame you for that.
While I am very happy for others, who are comfortable with the technical stuff to do this, if it's left to me, it won't happen.
I am a subscriber to the B&W Society of sound. They offer two albums every month (one classical and one other) in Alac 16/44, flac 16/44 and hirez flac 24/96 or 24/48 or 24/16.
Initially I always downloaded the hirez version but have stopped some time ago opting for 16/44 alac instead as I do not see any benefit in higher rez and do not want to waste hd space.
Initially I always downloaded the hirez version but have stopped some time ago opting for 16/44 alac instead as I do not see any benefit in higher rez and do not want to waste hd space.
shadders
Well-known member
fr0g said:
Hi,
Thinking a bit more about what you have done - you have taken a FFT of both the original 16bit and downsampled "24bit to 16bit" files.
The result is that the power density of the 0Hz to 17.5kHz spectrum (i assume power - could be voltage) difference is 90dB.
Did you normalise the calculation at all ?.
They do seem to have minimal differences in power density spectrum (90dB is 0.003% difference using voltage density), but this does not mean that they are exactly the same, sample for sample - in fact, the extra energy above 17.5kHz shows this.
All this means is that the spectrum of the signal for both recordings is closely matched.
I suppose that, if you believe that higher frequency signals can impact the lower frequency experience - then the extra high frequency content is important.
Regards,
Shadders.
oldric_naubhoff said:
Not according to this:
http://www.audiomisc.co.uk/ddd/DirtyDigitalDelusions.html
http://www.audiomisc.co.uk/inadither/Page1.html
Here's an excellent educational video:
http://xiph.org/video/vid2.shtml
oldric_naubhoff said:see below graphical representation of 1kHz sinewave recorded at -90dB referenced to full scale. this is only 6dB above resolution limit for 16 bit format and some 50dB above resolution limit for 24 bit format. first quantized with 16 bit data:
I was puzzled by this for a while and was wondering if the solution was dither. It seems my hunch was correct:
http://www.audiomisc.co.uk/inadither/Page2.html
shadders said:
Eh?
Just how do higher frequency signals impact the lower frequency experience?
Most posters on this forum will have sufficiently worn out hearing apparatus that they will struggle to hear anything above 15KHz, leave alone 17.5KHz, or are you postulating some form of intermodulation distortion?
shadders
Well-known member
andyjm said:
Hi,
I did say "if you believe that higher frequency signals can impact the lower frequency experience".
In the replay chain a loudpeaker driver may be affected by this ?.
I don't know - i have seen sttaements that you may not be able to hear the the frequency, but may be able to her the effect it has on the other content. I did not take much notice of the explanation.
Regards,
Shadders.
busb
Well-known member
I have read some of the technical papers linked to. The one on a design brief for a new DAC was fascinating (the bits I understood - I'm no expert on Complex planes & filter design). This paper also suggested that the case for 24/96 was quite strong - not for the extra dynamic range (bit depth) or higher sampling rate (giving an increase in f response) but being able to relax filter design as being the most interesting issue raised (another design brief was low latency which I haven't really thought about). Some conclude that there is mileage in the final mix we buy being 24/96. Others still feel that although there are measurable differences for 24/96, they are essentially inaudible in a final format for consumers, punters or audiophiles! The paper also pointed out that pre-echo will be more obvious than post-echo because it doesn’t happen in nature so different filters have subtle effects.
The idea that any difference between 16/44.1 & 24/96 being provable is interesting & has highlighted the fact that many music companies have been less than honest by producing high bit rate copies with better mastering! It seems a pretty sound method to down-sample a 24/96 file to 16/44.1 at home, create a null copy that should be essentially silent (if there is indeed no difference) then compare the copy to the original, blind or otherwise. There are a couple of reasons why the copy could sound poorer unrelated to the bit depth - the down-sampling needs to be done without just truncating (throwing away) the least significant 8 bits. Poorer DACs may sound inherently better at 24/96 due to filter design being easier than for 44.1! So a couple of potential issues need to be considered before drawing conclusions
.
We have to be a little more careful how we interpret compression & DR. If we measure the dynamic range of a particular recording that comes out as being say a max of 20dB - that number refers to the total envelope at any moment - not all sounds below 20dB will be masked so many will still be heard - echoes being an example. The judicious use of compression to is also necessary to balance out differing strands in a composition. Voices naturally have a higher DR compared to instruments. My beef with compression is not so much the reduction of DR but the horrid artefacts such as volume pumping (audible amplitude distortion) that modulate other sounds in a mix when applied without care. The real issue for me remains recording quality rather than bit depth.
The idea that any difference between 16/44.1 & 24/96 being provable is interesting & has highlighted the fact that many music companies have been less than honest by producing high bit rate copies with better mastering! It seems a pretty sound method to down-sample a 24/96 file to 16/44.1 at home, create a null copy that should be essentially silent (if there is indeed no difference) then compare the copy to the original, blind or otherwise. There are a couple of reasons why the copy could sound poorer unrelated to the bit depth - the down-sampling needs to be done without just truncating (throwing away) the least significant 8 bits. Poorer DACs may sound inherently better at 24/96 due to filter design being easier than for 44.1! So a couple of potential issues need to be considered before drawing conclusions
.
We have to be a little more careful how we interpret compression & DR. If we measure the dynamic range of a particular recording that comes out as being say a max of 20dB - that number refers to the total envelope at any moment - not all sounds below 20dB will be masked so many will still be heard - echoes being an example. The judicious use of compression to is also necessary to balance out differing strands in a composition. Voices naturally have a higher DR compared to instruments. My beef with compression is not so much the reduction of DR but the horrid artefacts such as volume pumping (audible amplitude distortion) that modulate other sounds in a mix when applied without care. The real issue for me remains recording quality rather than bit depth.
We visited a science park yesterday,
one demonstration was to sit with headphones and adjust the frequence of a sine signal from 10 Hz - 30 kHz.
It was fun to sit and listen, adjust, and be convinced that 20Hz - 18kHz is more than enough as frequence interval.
one demonstration was to sit with headphones and adjust the frequence of a sine signal from 10 Hz - 30 kHz.
It was fun to sit and listen, adjust, and be convinced that 20Hz - 18kHz is more than enough as frequence interval.
Ok.
For anyone still following this thread. I downloaded the CD quality flac (it's only 1.5 euro so sod it).
Analysis was difficult as the start is slightly delayed. So I was forced to align the tracks first, which involves a bit of faffing in Audacity.
But as I can go and look at sample level, I managed it.
The null test was interesting.
First of all, to the naked eye, looking at both spectrums, the quality is equal. ie they look the same. Both well recorded, no clipping. They "looked" like the same recording. My guess is that they do come from the same master....however...
A null test shows they aren't, quite the same. Some of the difference file is silent, but there is a lot of voice and some instrument, suggesting the timing is slightly different. ie The 2 original files not exactly the same. My guess is that one has been "amended" a little.
So, they are different..It is entirely probable therefore that the 16 bit CD quality FLAC will sound "different" to the 24 bit file when you listen.
But just to reiterate, the 24 bit when resampled to 16 bit is audibly exactly the same as the 24 bit version (unless you believe the idea that very low volume high frequency sounds can somehow alter the sounds that are in the audible range).
Draw your own conclusions.
For anyone still following this thread. I downloaded the CD quality flac (it's only 1.5 euro so sod it).
Analysis was difficult as the start is slightly delayed. So I was forced to align the tracks first, which involves a bit of faffing in Audacity.
But as I can go and look at sample level, I managed it.
The null test was interesting.
First of all, to the naked eye, looking at both spectrums, the quality is equal. ie they look the same. Both well recorded, no clipping. They "looked" like the same recording. My guess is that they do come from the same master....however...
A null test shows they aren't, quite the same. Some of the difference file is silent, but there is a lot of voice and some instrument, suggesting the timing is slightly different. ie The 2 original files not exactly the same. My guess is that one has been "amended" a little.
So, they are different..It is entirely probable therefore that the 16 bit CD quality FLAC will sound "different" to the 24 bit file when you listen.
But just to reiterate, the 24 bit when resampled to 16 bit is audibly exactly the same as the 24 bit version (unless you believe the idea that very low volume high frequency sounds can somehow alter the sounds that are in the audible range).
Draw your own conclusions.
fr0g said:
Err... that there's some audio detail in the 24-bit file that isn't in the 16-bit? Why automatically assume the timing is different? Is it because the latest comparison you've made might just demonstrate the opposite to what you wanted it to?
shadders
Well-known member
fr0g said:
Hi,
I think we need more information - such as what was the sample rate of the 16bit FLAC ? and the 24bit version too ?
Regards,
Shadders.
spiny norman said:
LOL no.
The audio detail was the same. Just timed slightly differently..
I was "expecting" there to be a difference.
As explained. If I downsample to 16 bit, then compare, there is NO difference (audibly).
What I was trying to prove is that the 16 bit file as supplied "IS" different, which it is.
The only surprise was how little different it was, and in what way.
The differences that showed up were absolutely nothing to do with bit depth or sample rate.
Er, that's in your opinion. I mean, are you an expert in this, or like the rest lf us, an amateur?
shadders
Well-known member
fr0g said:
Hi,
Can you explain clearly the process you implemented ?
If you use a computer program to resample a file from 24bit to 16bit, the algorithm used will produce one conversion, another program will produce another conversion (possibly).
One program may set a sample LSB to a 0 (zero) whilst another program will set this LSB to a 1 (one) - when calculating the 16bit value.
Are these both 48kHz sample rate files ?
Regards,
Shadders.
shadders said:
In my first test (documented earlier in the thread) I used a 24/96 recording. Downsampled to 16/44.1 in Audacity. No special settings.
I made a null file which was essentially silent, but contained some high frequency noise at around -80 dB
In the last 2 tests I took a recommended "good HD recording", which was 24/48
I downsampled to 16/48 and did another null test...similar result
The last test was between the 24/48 recording and the 16/44.1 recording from the same site. These were ever so slightly different (timing).
What it demonstrates to me is that the HD file is not necessary for playback and that the differing formats use different recordings, or at least recordings that have been made different.
Another thing it demonstrates very well is Nyquist Shannon...
The waveforms in the human audible range in the 96 KHz sample rate file, when resampled to 44.1 were in every aspect, identical, other than some noise well below the level you would hear in dB terms, and at around 18 KHz and up.
shadders
Well-known member
fr0g said:
Hi,
The first test i recall, -90dB across the audio band to 17.5kHz where from here on there were differences up to -76dB.
This was just a comparison of the spectral differences, not the actual waveform differences. You have also downsampled to a different sample frequency where the two frequencies have no common factor - could be termed relatively prime to one another.
For the second test - 24bit/48kHz and resampled to 16bit/48kHz - was this a null file listened to or frequency domain comparison - spectrum ?
The third test (last test stated by yourself) downsampled from 24bit/48kHz to 16bit/44.1kHz. I recall that you listened to the difference (null) file. You indicate timings - but this may be due to the downsampling to a sampling frequency that has not common factor to the original sampling frequency.
In you first (96kHz to 44.1kHz) and third (48kHz to 44.1kHz) test the sampling rate differences (no common factor) will introduce artefacts - which i would expect are dependent upon the algorithm used. I have not studied this aspect and the effects of this type of downsampling - hence assumptions.
What you seem to be inferring is that the downsampling and the subsequent frequency/spectrum comparison provides evidence that you cannnot hear the differences between the two.
The two spectrums of the files compared are not the same else the differences would not be -90dB etc., but would be infact, {minus infinity dB} in the perfect world, but Audacity (assumed program that you used) processes in a specific way and is limited by the algorithm, and internal number representation etc., and is limited by the accuracy of the word depth per sample (i assume this at this time, not studied).
I think your tests have shown that you do not hear differences between standard 16bit 44.1kHz sources and an HD source (24bit/96kHz) hence it is not worth yourself paying extra for a suggested better experience, that you will not receive.
I have some Pure Audio blu-ray discs - i have not compared to CD as i do not have the original CD version - so i may be spending extra on something i too will not benefit from. The discs do sound good, but again, placebo affect may be the issue here.
Regards,
Shadders.
Facebook
Twitter
Instagram