« Back

Analysis of submitted ATRAC SP (Type-R) recordings

1. Collection of submitted recordings

All submitted recordings were processed with time warping algorithm and listed in the table below:

Author of
submitted
recording

Difference levels (dB) of time warped SE samples
(Left and Right channels)

Download
BAH BAS CST FMS GLK HRP LOB MOF QRT
L R L R L R L R L R L R L R L R L R
Lee1 -38.1 -38.7 -41.1 -40.0 -19.4 -19.7 -35.0 -34.9 -30.3 -30.7 -19.9 -19.9 -26.2 -22.6 -27.3 -27.0 -37.1 -36.2  
Lee2 -38.5 -39.2 -40.9 -39.4 -19.3 -19.8 -34.7 -34.9 -31.6 -31.2 -19.8 -20.1 -25.7 -22.0 -26.6 -26.7 -36.8 -36.3 [link removed]
Lee3 -38.5 -39.0 -41.3 -40.5 -19.5 -19.5 -34.8 -35.2 -31.1 -31.2 -19.7 -19.8 -26.0 -22.3 -26.7 -26.6 -36.6 -36.2  
                                       
Numinos1 -37.3 -37.7 -39.2 -38.8 -18.9 -19.2 -34.2 -34.3 -29.5 -29.3 -19.3 -19.5 -25.1 -21.7 -26.0 -26.0 -36.3 -35.5  
Numinos2 -37.8 -38.0 -39.8 -38.9 -19.2 -19.1 -34.6 -34.4 -30.1 -29.9 -18.6 -19.7 -25.0 -21.8 -27.0 -26.6 -36.3 -35.5  
Numinos3 -37.6 -38.3 -39.4 -38.7 -18.6 -19.3 -34.5 -34.6 -29.2 -28.8 -19.7 -19.3 -25.9 -22.5 -27.0 -26.7 -36.4 -35.6 [link removed]
                                       
MikeRofone1 -31.6 -31.7 -32.8 -32.7 -18.1 -18.4 -30.9 -30.9 -30.0 -28.7 -19.3 -19.4 -24.1 -21.2 -25.1 -24.9 -31.7 -31.5 [link removed]
MikeRofone2 -31.6 -31.7 -32.7 -32.6 -18.3 -18.6 -30.9 -30.8 -26.5 -28.1 -19.1 -19.5 -24.5 -21.3 -25.1 -24.9 -31.6 -31.5  
MikeRofone3 -31.6 -31.7 -32.9 -32.7 -18.2 -18.5 -30.9 -30.7 -25.8 -24.9 -19.2 -19.3 -23.8 -20.5 -25.1 -24.4 -31.7 -31.4  
                                       
MikeRofone21 -38.6 -38.9 -41.1 -39.9 -19.2 -19.4 -34.6 -34.8 -33.7 -31.9 -20.1 -20.5 -25.5 -22.2 -27.2 -26.7 -36.9 -36.4  
MikeRofone22 -38.5 -38.8 -40.6 -39.4 -19.4 -19.6 -34.5 -34.7 -30.0 -30.4 -19.8 -19.9 -25.8 -22.2 -26.7 -26.5 -36.9 -36.3 [link removed]
MikeRofone23 -38.2 -39.0 -41.3 -39.5 -19.3 -19.4 -34.7 -34.5 -27.1 -25.9 -19.8 -20.0 -24.8 -21.5 -27.0 -26.8 -36.8 -36.2  

 

Notes:

  1. Difference level shows how much does the warped signal differ from reference one.
       -Infinity(dB) means identical signal shapes,
        0(dB) – completely different shapes.
  2. As time warping procedure is computationally extensive fast version of the algorithm was used for processing of all submitted recordings. The candidate chosen for inclusion to SE rating system will be processed with the highest possible accuracy. Accurate time warping gives slightly lower values of Difference levels (around 1dB lower) and lower contamination of processed signal with time warping artifacts (noticeable in the above difference signals as amplitude modulation).
  3. Test file used for recording consists of three identical groups of SE test samples. Difference levels are indicated for all of them. Time warped versions are uploaded for one of them only as they are almost similar.

 

2. Cluster analysis

Here is graphical representation of Diff. levels from the table above:

Difference levels of ATRAC SP recordings submitted to SoundExpert

 

All recordings except MikeRofone1-3 look consistent with each other. However the graph doesn't show how similar they are. Cluster analysis is more informative in this sense:

Cluster analysis of ATRAC SP recordings submitted to SoundExpert

This dendrogramm shows how the recordings break up into groups according to similarity to each other. The higher the |¯|-shaped lines the less similar connected recordings. In our analysis all recordings could be divided into three groups:

  1. Lee1-3 and MikeRofone22 are most similar. This is not surprising because Lee did some research to find the setup which provides the lowest number of errors while recording-playback-recording on his equipment. MikeRofone22 also hit the group and MikeRofone21 is the second closest. However MikeRofone21-23 recordings are less similar to each other. 
  2. Numinos1-3 recordings are also less similar to each other but pretty close to the first group.
  3. MikeRofone1-3 recordings differ greatly from all others. They have to be excluded from further analysis.

 

3. Selection of best samples

Recordings from group#1 - Lee1-3 and MikeRofone22 plus MikeRofone21,23 - were time warped with higher accuracy. Results are below: 

Author of
submitted
recording
Difference levels (dB) of time warped SE samples
(Left and Right channels mixed)
Download
DIFF signals
BAH BAS CST FMS GLK HRP LOB MOF QRT
Lee1 -39.42 -43.25 -22.11 -37.44 -33.60 -21.10 -27.12 -30.81 -37.44 [link removed] 
Lee2 -39.88 -42.84 -22.06 -37.65 -33.45 -21.01 -26.85 -31.33 -37.40 [link removed] 
Lee3 -39.80 -43.69 -22.05 -37.57 -34.13 -20.83 -26.58 -31.06 -37.21 [link removed] 
                     
MikeRofone21 -39.76 -43.22 -22.01 -37.52 -34.01 -21.56 -26.40 -31.19 -37.36 [link removed] 
MikeRofone22 -39.65 -42.73 -21.97 -37.34 -33.39 -21.48 -26.95 -30.98 -37.36 [link removed] 
MikeRofone23 -39.64 -43.09 -21.82 -37.40 -32.46 -21.54 -26.50 -31.16 -37.27 [link removed] 

 

Note: It turned out that Diff. level of mixed Left and Right channels is more sensitive parameter because it also accounts phase shifts between channels. From now on it will be used instead of separate Diff. levels.

Visual representation of the values: 

Difference levels of selected ATRAC SP recordings

 

Audibly similar type of artifacts in diff signals and close values of Diff. levels clearly show that all recordings belong to Type-R compression. The only difference between them is number of recording errors. In other words degradation of the signals caused by two factors - ATRAC compression and recording errors.

Unfortunately each author's recording has only three observations for each sound sample. This doesn't allow us to perform analysis of variance which would reveal distribution type of those observations. So we assume that it is one-sided normal distribution because it is highly unlikely that recording errors could improve Diff. levels over those conditioned by ATRAC. If so we could chose for each sound sample the recording that has the lowest Diff. level of warped signal, which means that the recording is less affected by recording errors. These lowest values are marked with green in the table. 

The selected recordings were chosen as the best representatives of ATRAC SP (Type-R) encodings of nine SoundExpert samples. They were then time-warped with accuracy below -100dB - thus hiding artifacts of time-warping beyond resolution of 16 bit audio. Resulting Diff. levels, diff. signals and time-warped samples are in the table:

Author of
submitted
recording
Difference levels (dB) of time warped SE samples
(Left and Right channels mixed)
Download
BAH BAS CST FMS GLK HRP LOB MOF QRT
Lee1     -22.28       -27.10   -37.46 diff [link removed]
warp [link removed]
Lee2 -39.90     -37.69       -31.38  
Lee3   -43.82     -34.16        
MikeRofone21           -21.62      

 

 

4. Contributors and equipment

SoundExpert would like to thank people who recorded SE test file on their equipment:

Lee Keywood (www.leekeywood.co.uk)

Recording chain: SE test file(.wav) ==> PC(Linux) ==S/PDIF==> Sony MZ-R410(SP,Type-R) ==MD-disc==> Technics SJ-MD100 ==S/PDIF==> PC(Linux) ==> Lee1-3(.wav)

Numinos from Minidiscforum.de (forum post)

Recording chain: SE test file(.wav) ==> PC ==USB-S/PDIF==> Sony MZ-NH600(SP,Type-R) ==MD-disc==> Kenwood DM-3090 ==S/PDIF==> Sony MZ-NH600(PCM) ==upload as PCM through SonicStage==> PC ==> Numinos1-3(.wav)

Mike Rofone from Audio T-board (forum post)

  • Recording chain: SE test file(.wav) ==> Audio CD ==> Pioneer CD player ==Toslink==> Sony MZ-RH1(SP,Type-R) ==upload as Hi-SP through SonicStage==> PC ==> MikeRofone1-3(.wav)
  • Recording chain: SE test file(.wav) ==> Audio CD ==> Pioneer CD player ==Toslink==> Sony MZ-RH1(SP,Type-R) ==upload as PCM through SonicStage==> PC ==> MikeRofone21-23(.wav)
Comments
No comments yet. Be the first.