Many have claimed that wav files obtained from ripping on different CD drives sound different. I had always found this very intriguing. At the RMAF in the EA room I saw Steve demonstrate that these files do sound different. He had played three files and they all sounded different.
A couple of days back I requested Steve to send me these three wav files and he obliged. I wrote a small program to convert these wave files into a more readable text format. The wav file format is very simple and I used this link to understand it.
https://ccrma.stanford.edu/courses/422/projects/WaveFormat/The converted text files looked as below...
-------------------------------------------
ChunkID = 0x46464952
ChunkSize = 72455018
Format = 0x45564157
Subchunk1ID = 0x20746d66
Subchunk1Size = 16
AudioFormat = 1
NumChannels = 2
SampleRate = 44100
ByteRate = 176400
BlockAlign = 4
BitsPerSample = 16
Subchunk2ID = 0x61746164
Subchunk2Size = 70795200
Stereo Data Samples (L R) follow....
0 0
0 0
0 0
0 0
0 0
0 0
0 0
-1 0
-1 -1
---
---
---
7739 12515
9950 11382
9349 9173
6346 6270
2634 3135
142 141
-1054 -2347
-1849 -4442
-2844 -6458
-3989 -8388
---
---
---
-1 0
0 0
1 1
-1 -1
0 0
0 0
0 0
0 0
0 0
0 0
--------------------------------------------------------
Basically it contains 44 bytes of header and then the signed stereo samples. I found that the only field which was different in these headers across the three files was the "ChunkSize" field, which is effectively the file size. All the three files had different number of zero samples in the beginning and end of the data samples. I verified that once you delete some of these zero samples in the files to align them, they compared perfectly for the remaining data samples.
I feel it is only number of leading zero samples which make a difference in the sound quality. After all the DAC cannot know what the tail samples are going to be upfront. Perhaps the adequate number of zero samples at the start of the song, help to reset the D/A chip. Most of the D/A chips have delta-sigma architecture. There are highly recursive integrators in the delta-sigma modulators. The junk residues in these integrators at the beginning of the song can have a effect on the rest of the song. More ever perhaps different junk residues in the two stereo channels may affect the imaging too.
If anybody needs a primer on delta-sigma modulation (I did), this site might be useful...
http://www.beis.de/Elektronik/DeltaSigma/DeltaSigma.htmlI looked at the leading zero samples in the three wav file A,B,C. The file C had no leading zero samples. File B had around 100 more leading zero samples than file A. Using the above argument I speculated that file C should sound the worst and file B should sound the best. File A should sound worse than B but very close. Steve confirmed that was actually the case!
I actually think that the file B can sound still better if the number of leading zero samples are increased further. After all file A has around 450 of those while B has only a hundred more. If the number of leading zero samples really make a difference in sound quality, then even the media player can take the responsibility of sending a burst of zero samples at the start of the song. Perhaps Amarra can implement this.