Head-Fi.org › Forums › Equipment Forums › Sound Science › 96/24 Vs. 44.1/16 Vs. MP3 320
New Posts  All Forums:Forum Nav:

96/24 Vs. 44.1/16 Vs. MP3 320

post #1 of 5
Thread Starter 

Dear all,


I wake up very early this morning, and as my wife did not want to speak with me at those hours, I had to occupy my time with a little experiment.


First aim was to find a way to verify if there is a way to verify that the process I am using to rip my CD's was properly implemented. I am using EAC with Accuraterip and an external converter for translating the files to Apple Lossless (convenience). I was a little afraid that the external converter wasn't good enough, so I read on Internet how to verify it:


1) Install Audacity

2) Import Audio File 1 (m4a file)

3) Import Audio File 2 (wav file direct from EAC)

4) Invert the first audio file

5) Mix and Render both

6) Look for any audio signal left after the process.


I was happy to verify that the process was working perfectly, and that any signal appear after the mix, as expected.


Then, I though it could be interesting to compress one audiofile from HDTracks to MP3 320 and see any difference, since in my kind of blind test I could not find any. Result: Yes, there are clearly audible differences between the files. 


The software I used to convert from 96/24 to mp3 is called "Switch Sound File Converter". I do not know what algoriythm it uses, only that it is VBR.


I use the same software to downsample the 96 file to lossless 44.1/16, and using the same procedure describe above to compare the differences with Audacity, I got nothing to hear (there was information, but it was at a very low volume); if I use the lowpass filter, even as low as 5kHz, I still have audio in there. I cannot listen to it because it is really low volume, but I though any difference would be above 22 kHz.


Can anybody please explain it to me? Why I have signal as such "low" frecuency?


Best Regards,



Edited by daniel0407 - 2/1/14 at 3:14am
post #2 of 5

What method did you use for the blind test? Was it repeatable?  Did you volume match first?


Daniel - if you're open to using foobar - here's a quick 'beginner's guide' to setting up their abx tool.  You can apply replay gain as well - so that you make sure the two tracks are volume matched.  Then you've got grounds for a good test.


Link : http://www.head-fi.org/t/655879/setting-up-an-abx-test-simple-guide-to-ripping-tagging-transcoding


Sorry - I can't answer your low pass question - but there are guys here who will be able to.

post #3 of 5
When you are converting to a lower bit depth, quantization error may add noise to the signal which includes lower frequencies.

As for mp3 you can try using an up-to-date mp3 encoder.
Edited by higbvuyb - 2/1/14 at 7:50am
post #4 of 5
Thread Starter 

Dear all,


I realise I made two mistakes:


1) When applying the low pass filter, I did not use enough "roll off" per octave, only 6dB. Now I applied 48dB at 20kHz, but I am still getting audio signal below this frequency. I amplified the remaining signal 45 dB, and I could hear the song, with bad quality, lot of hiss, but a many many frequencies were there. I am very confuse with it. Could somebody please explain why a very big portion of the song is still there, after "nulling" it using the two different sampling rates? Volume looks the same for the two files. Could it be that there is a very very little tiny difference in volume created by the software, when downsampling the file? If so, how can one avoid this and test the difference properly?


2) When comparing MP3 vs HD File, it could be that both files does not have the same volume, which could be the reason behind this big quality difference. As mentioned, could be that the mp3 algorithm is modifying the volume level of the resulting audio file. Is there a way to reduce the volume of the MP3 accurately to match that of the HD File?


Thanks again,



Edited by daniel0407 - 2/1/14 at 10:16am
post #5 of 5

It is common for MP3 encoding/decoding to slightly decrease the volume. It will also normally produce a relatively large difference signal, but the noise is psycho-acoustically "optimized" so that it is masked well by the original signal.


If you get too much difference with the resampler, it may be due to one of these reasons:

- perhaps it is a bad resampler. :normal_smile : Try the SoX sample rate converter, or the one from the third link in my signature for accurate results

- there may be a small volume difference. Even if the volume is "off" by only 0.01 dB, which is inaudible, you get a -60 dB difference signal

- there could be a slight delay - even if only a fraction of a sample - in the output of the resampler. If this is the case, the difference will sound highpass filtered (increasing by 6 dB/octave over the entire band)

Edit: also, if the resampler uses a minimum phase filter, you will not get an accurate null difference.

Edited by stv014 - 2/2/14 at 11:29am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › 96/24 Vs. 44.1/16 Vs. MP3 320