Performace Test: aacenc32.dll vs. AAC-CLI-Encoder

09.05.2006 by Naito

lousy translation made by Redfox (no, I don't have an english spell-cecker installed on my machine)

Last weekend I (Naito) showed my AMD-Duron 800 how it feels to be a Calculator. Alltogetter, it had to encode 32h and 15min Audio into AAC. After Neros release of the AAC-CLI-Encoder I decided to make some exausive testing. The following Spreedsheds containing the results.

1. Testsystem

  • AMD Duron @ 832MHz with 512MB SD-RAM
  • Windows XP Professional incl. SP 2 (no Patches)
  • Nero aacenc32.dll 3.2.0.30 (from Nero 6.6.1.4)
  • Nero AAC-CLI-Encoder 01.05.2006
  • BeSweet 1.51 Beta31
  • foobar2000 0.9.1 with AC3-Plugin 0.7 (foo_ac3.dll)

    • 2. Test Conditions

  • I used BeSweet to transcode with aaenc32.dll.
    The Commandline locked like that:
    "BeSweet.exe -core( -input AC3.ac3 -output Art-Profil.mp4 -logfile Art-Profil.log ) -bsn( -6chnew -vbr_Profil -codecquality_high -aacprofile_Art)"
    'Art' means LC or HE. Profil reachs from Tape till Transcoding.
  • For transcoding of VBR with the AAC-CLI-Encoder I useed foobar2000.
    - The commandline for 'forsed' LC or HE locked like this:
    HE (force): "-ignorelength -he -q 0.xx -if - -of %d"
    LC (force): "-ignorelength -lc -q 0.xx -if - -of %d"
    - The commandline for 'free' VBR locked like this:
    VBR (free): "-ignorelength -q 0.xx -if - -of %d"
    Replace 'xx' with the certain values. The "-" tells foobar2000 to use a pipe.
  • For the ABR-Transcoding I converted AC3 with BeSweet into DD-WAV and feeded the AAC-CLI-Encoder with a Batch-File.
  • Für 5.1@2.0 I used the DSP in foobar2000 activated and the same Commandlineparameter like for 5.1
  • The sourcefile was the german soundtrack of „Ghost in the Shell ~ Stand Alone Complex – Episode 01“.

    • sourcefile
    bitrate = 448
    codec = ATSC A/52
    channels = 6
    samplerate = 48000
    ----------
    72388608 samples @ 48000Hz = 25:08.096 (Length)
    File size: 84 453 376 bytes = 80,5 MB

    3.1 Results: 5.1-Test (the interesting stuff is at the bottom)

    units
    		 aacenc32.dll 3.2.0.30 CLI (free) CLI (force) CLI
    Profil  HE LC   LC/HE VBR  Hybrid VBR HE LC LC/HE ABR 2pass Abw
    kbps
    		 Tape
    		 109
    		 155
    		 1,42
    		 0.10
    		 n. m.
    		 0.10
    		 81
    		 165
    		 2,04
    		 100
    		 104
    		 1,04%
    %
    		24,3%
    		 34,6%



    		18,1%
    		 36,8%
    		HE
    		 23,2%
    size
    		19,74
    		 28,23



    		14,68
    		 29,98

    		18,80
    speed
    		0,85
    		 084



    		1,41
    		 0,88

    		0,82
    kbps Radio
    		 129
    		 203
    		 1,57
    		 0.15
    		 101
    		 0.15
    		 117
    		 224
    		 1,91
    		 125
    		 126
    		 1,01%
    %
    		28,8%
    		 45,3%
    		HE
    		 22,5%
    		26,1%
    		 50,0%
    		HE
    		 28,1%
    size
    		23,33
    		 36,80

    		18,33
    		21,16
    		 40,52

    		22,82
    speed
    		0,85
    		 0,80

    		1,13
    		1,23
    		 0,87

    		0,81
    kbps Internet
    		 140
    		 307
    		 2,19
    		 0.20
    		 145
    		 0.20
    		 153
    		 286
    		 1,87
    		 150
    		 150
    		 1,00%
    %
    		31,3%
    		 68,5%
    		HE
    		 32,4%
    		34,2%
    		 63,8%
    		HE
    		 33,5%
    size
    		25,34
    		 55,44

    		26,19
    		27,69 51,78

    		27,10
    speed
    		0,85
    		 0,72

    		1,18
    		1,19
    		 0,82

    		0,74
    kbps
    		 Streaming
    		 172
    		 381
    		 2,22
    		 0.25
    		 190
    		 0.25
    		 191
    		 347
    		 1,82
    		 175
    		 176
    		 1,01%
    %
    		38,4%
    		 85,0%
    		HE
    		 42,4%
    		42,6%
    		 77,5%
    		HE
    		 39,3%
    size
    		31,10
    		 68,75

    		34,25
    		34,52
    		 62,57

    		31,82
    speed
    		0,81
    		 0,68

    		1,25
    		1,16
    		 0,82

    		0,73
    kbps Normal
    		 190
    		 485
    		 2,55
    		 0.30
    		 234
    		 0.30
    		 228
    		 405
    		 1,78
    		 200
    		 198
    		 0,99%
    %
    		42,4%
    		 108,3%
    		HE
    		 52,2%
    		50,9%
    		 90,4%
    		HE
    		 44,2%
    size
    		34,31
    		 87,51

    		42,17
    		41,10
    		 73,17

    		35,78
    speed
    		0,79
    		 0,58

    		1,19
    		1,16
    		 0,84

    		0,73
    kbps Extreme
    		 273
    		 625
    		 2,29
    		 0.31
    		 263
    		 0.31
    		 235
    		 417
    		 1,77
    		 225
    		 219
    		 0,97%
    %
    		60,9%
    		 139,5%
    		LC
    		 58,7%
    		52,5%
    		 93,1%
    		HE
    		 48,9%
    size
    		49,26
    		 112,61

    		47,53
    		42,43
    		 75,27

    		0,32
    speed
    		0,76
    		 0,44

    		0,73
    		1,12
    		 0,81



    kbps Audiophile
    		 308
    		 646
    		 2,10
    		 0.35
    		 320
    		 0.35
    		 264
    		 470
    		 1,78
    		 250
    		 242
    		 0,97%
    %
    		68,8%
    		 144,2%
    		LC
    		 71,4%
    		58,9%
    		 104,9%
    		HE
    		 54,0%
    size
    		55,58
    		 116,36

    		57,82
    		47,64
    		 84,70

    		43,71
    speed
    		0,79
    		 0,46

    		0,82
    		1,32
    		 0,83

    		0,74
    kbps Transcoding
    		 339
    		 744
    		 2,19
    		 0.40
    		 390
    		 0.40
    		 299
    		 528
    		 1,77
    		 275
    		 274
    		 1,00%
    %
    		75,7%
    		 166,1%
    		LC
    		 87,1%
    		66,7%
    		 117,9%
    		LC
    		 61,2%
    size
    		61,00
    		 134,09

    		70,42
    		53,92
    		 95,19

    		49,61
    speed
    		0,77
    		 0,55

    		0,79
    		1,23
    		 0,84

    		0,50
    kbps



    		0.45
    		 465
    		 0.45
    		 334
    		 586
    		 1,75
    		 300
    		 293
    		 0,98%
    %



    		LC
    		 103,8%
    		74,6%
    		 130,8%
    		LC
    		 65,4%
    size




    		83,93
    		60,21
    		 105,57

    		52,91
    speed




    		0,81
    		1,26
    		 0,85

    		0,51
    kbps



    		0.49
    		 521
    		 0.49
    		 362
    		 633
    		 1,75
    		 325
    		 315
    		 0,97%
    %



    		LC
    		 116,3%
    		80,8%
    		 141,3%
    		LC
    		 70,3%
    size




    		93,92
    		65,26
    		 114,08

    		56,83
    speed




    		0,84
    		1,28
    		 0,81

    		0,51
    kbps



    		0.50
    		 535
    		 0.50
    		 n. m.
    		 644
    		350
    		 337 0,96%
    %



    		LC
    		 119,4%
    		n. m. 143,8%
    		LC
    		 75,2%
    size




    		96,39
    		n. m. 116,14

    		0,31
    speed




    		0,84
    		n. m. 0,84



    kbps









    		375
    		 348
    		 0,93%
    %









    		LC
    		 77,7%
    size










    		0,31
    speed












    kbps









    		400
    		 367
    		 0,92%
    %









    		LC
    		 81,9%
    size










    		0,31
    speed












    kbps









    		425
    		 389
    		 0,92%
    %









    		LC
    		 86,8%
    size










    		0,31
    speed












    kbps









    		450
    		 410
    		 0,91%
    %









    		LC
    		 91,5%
    size










    		0,31
    speed












    aprox. kbps
    		0,81
    		 0,63
    		 2,07
    		0,96
    		1,24
    		 0,84
    		 1,85
    		0,59
    standartdevation
    		0,04
    		 0,15
    		 0,38
    		0,20
    		0,09
    		 0,02
    		 0,10
    		0,20
    aprox. overhead
    		0,19%
    		 0,35%

    		0,26%
    		0,20%
    		 0,35%

    		0,28%
    		 0,97%
    The Deviation of the Bitrate is around 2%.
    0.00 <= 0.15 is HEv2; 0.16 <= 0.30 is HE; 0.31 <= 0.49 is HE+LC (?); 0.50 < 1.00 is LC
    ABR: 100kbps = 100.000bps

    Note that the standartdevation is the root from the vary, but Excel calculates them wrong. So better be carefull with those values. I didn't corected those results because they don't matter statisticly. The are just for your orientation. The avertage Overhad is calculated by the percentage part of the filesize minus the percentage of the bitrate(You can find the values in an attachment at the original posting)

    3.2 Results: 5.1@2.0

    units CLI (free) CLI (force)
    VBR Hybrid 5.1/2.0  VBR HE 5.1/2.0
    		 LC 5.1/2.0
    		 LC/HE
    kbps
    		 0.10
    		 24
    		 n. m.
    		 0.10
    		 22
    		 3,68
    		 40
    		 4,13
    		 1,82
    size (MB)
    		 4,50


    		4,03
    		 3,64
    		 7,46
    		 4,02
    speed
    		 4,51


    		3,41
    		2,54

    kbps
    		 0.20
    		 42
    		 3,45
    		 0.20
    		 44
    		 3,48
    		 79
    		 3,62
    		 1,80
    size (MB)
    		 7,64
    		 3,43

    		8,00
    		 3,46
    		 14,40
    		 3,60
    speed
    		 2,86


    		2,87
    		2,11

    kbps
    		 0.30
    		 69 3,39 0.30 68 3,35 113 3,58 1,66
    size (MB)
    		 12,60
    		 3,77

    		12,33
    		 3,33
    		 20,61
    		 3,55
    speed
    		 2,80


    		2,89
    		2,12

    kbps
    		 0.40
    		 109
    		 3,58
    		 0.40
    		 86
    		 3,48
    		 140
    		 3,77
    		 1,63
    size (MB)
    		 19,87
    		 3,54

    		15,59
    		 3,46
    		 25,41
    		 3,75
    speed
    		 2,16


    		2,97
    		2,12

    kbps
    		 0.50
    		 141
    		 3,79
    		 0,56
    		 102
    		 n.m.
    		 186
    		 3,46
    		 1,82
    size (MB)
    		 25,71
    		 3,75

    		18,51
    		33,69
    		 3,45
    speed
    		 2,07


    		2,83
    		2,10

    average (kbps)

    		3,55

    		3,50
    		3,71
    		 1,75
    standartdeviation(kbps)

    		0,18

    		0,14
    		0,26 0,09
    average (MB)

    		3,62

    		3,47
    		3,67
    standartdeviation(MB)

    		0,17

    		0,13
    		0,22

    4.Appraisal of Results

  • The optinization is most noticeable with LC at multichannel. While  3.2.0.0.30 LC is aprox. 2,1-times bigger then HE, the improvement with the CLI-Encoder is stronger noticeable (aprox. 1,85).
  • It is interesting that an 5.1 encode is aprox. 3,5 times bigger then an similer 2.0. This might indicate an insufficient channelcuppling.
  • HE is only posible till „-q 0.49“, with values above  Nero displays an error and refuses to encode.
  • The 'free'-VBR-Modus (-q 0.xx without  -lc or -he switch)is quite interesting. The bitrates here are very differend then with forced switchs (maybe internal encoder-optimasions around the psychacusic-model?) We have to use liesening-tests to figure out what modus offers the best Audio quality) But you shuldnt force the switchs without an good reason.

    • Bysides, if somebody asks hiself  „Why is he doing this?“. Plain simple, its fun for me.

    The attachment contains screenshots, grafics, spreadsheets, batch-files and logs. (foobar2000 was 'handmade')

    source+attachment.
    Zuletzt bearbeitet: 11.05.06