Help - Search - Members - Calendar
Full Version: Lame 3.97 beta 1 released
Hydrogenaudio Forums > Lossy Audio Compression > MP3 > MP3 - Tech
Pages: 1, 2, 3
guruboolez
Did the --athaa-sensitivity switch disapear?
LAME 3.97b1 doesn't start with this switch, whereas the 3.97a12 using the same commandline is working perfectly. Is it intended?
guruboolez
QUOTE(john33 @ Sep 17 2005, 10:45 PM)
Always a difficult one, this, as it does very much depend upon the material used. However, if one can generalise, I think that it should be revised as follows:
CODE

Quality Setting                   Indicated Bitrate
 10  (V9)                                   64
 20  (V8)                                   92
 30  (V7)                                  112
 40  (V6)                                  128
 50  (V5)                                  140
 60  (V4)                                  160
 70  (V3)                                  176
 80  (V2)                                  200
 90  (V1)                                  224
 100 (V0)                                  240

Anyone else have a view on this?
*


I'm building a complete table based on a representative [of my own collection] set of 150 classical files (and only classical!). My computer is still busy; I think I will post the result tomorrow.
robert
QUOTE(guruboolez @ Sep 19 2005, 10:36 AM)
Did the --athaa-sensitivity switch disapear?
LAME 3.97b1 doesn't start with this switch, whereas the 3.97a12 using the same commandline is working perfectly. Is it intended?
*


the ath switches are available in alpha builds for developing and tuning. If you think you need the sensitivity switch to undo/tune presets, then the presets should be changed.
Gabriel
Experimental switches are also available when binary is compiled in debug mode.
[proxima]
In the past, with v3.96.1, the athaa-sensitivity 1 setting showed a great improvement with certain passages reducing ringing/warblig problems with very limited bitrate increase (at least for certain genres).
I didn't test latest 3.97 alhpa but according to guruboolez intensive testing, those problems are reduced to acceptable levels now. So, i suppose, he himself would not recommend using athaa switches with the new 3.97 version. If not and if 3.97a12 outputs bit identical results with 3.97b1 i can eventually test some samples.
guruboolez
QUOTE(robert @ Sep 19 2005, 09:59 AM)
QUOTE(guruboolez @ Sep 19 2005, 10:36 AM)
Did the --athaa-sensitivity switch disapear?
LAME 3.97b1 doesn't start with this switch, whereas the 3.97a12 using the same commandline is working perfectly. Is it intended?
*


the ath switches are available in alpha builds for developing and tuning. If you think you need the sensitivity switch to undo/tune presets, then the presets should be changed.
*


The quality impact of --athaa-sensitivity 1 (with -V5) is variable. I noticed difference in bitrate up to 20 kbps. Difference in quality is very important too - especially after MP3gain process (useful for portable IMO).
For me, it's a pity that this switch isn't available anymore. crying.gif
robert
is this switch only useful with -V5 preset, and maybe below?
would you always use it in combination with -V5 (saw some recommendation elsewhere)?

if not always, what kind of samples are affected?
Gabriel
QUOTE
The quality impact of --athaa-sensitivity 1 (with -V5) is variable. I noticed difference in bitrate up to 20 kbps. Difference in quality is very important too - especially after MP3gain process (useful for portable IMO).
For me, it's a pity that this switch isn't available anymore.

I'd suggest --athlower or --scale in such cases.
Assault
Please let me repeat AndyMutz's question:

QUOTE
when using --vbr-new the -q switch produces exact same results, no matter if i use q0,q1,q2,q3 or q4. is that intended?


Does anyone know anything about this? Thanks for your help!
guruboolez
QUOTE(robert @ Sep 19 2005, 11:25 AM)
is this switch only useful with -V5 preset, and maybe below?
would you always use it in combination with -V5 (saw some recommendation elsewhere)?

if not always, what kind of samples are affected?
*


I haven't tried the effect of the switch with VBR mode below -V5.
I always add this command to all -V5 encodings. On average the bitrate is increased by 6 kbps IIRC.
The impact is really audible on low volume sample/part of the sample. It becomes dramatic after MP3gain with such samples. But the --ahtaa-sensitivity 1 tool can't fully solve the extreme ringing issues occuring with very quiet samples. The quality gain is nevertheless important, and I would rather make --athaa-sensitivity 1 the default mode if you really want to remove this tool from the public version of LAME.



P.S. I recently made some RG statistics with a part of my classical library (11000 files were involved). ~1050 files had a track gain > +10 dB! On such tracks, there's serious ATH issues with LAME VBR with -V5. -V4 is by default much better.
[proxima]
I can speak only for 3.96.1 and early 3.97 alphas, i have to test the lastest version to be sure. --athaa-sensitivity 1 is useful with low/mid bitrate VBR settings. With -V5 the quality impact with some quite tracks in impressive but not resolutive with extreme samples. I remember i did tests even at -V4 and the same switch proved to be useful to reduce ringing. With -V3 and below i believe that the quality gain is trascurable. That said, i never mp3gained my encodings for these tests but i can't exclude that with very high corrections (or high volume listening) artifacts could arise even with others VBR settings.

@robert: if you search for athaa sensitivity in the forums, you'll find some tests and samples affected but i believe that these cases are not so unusual, expecially with certain genres.

@gabriel: --athaa-sensitivity corrects well some issues with limitate bitrate increase. --athlower will certainly help in the same cases but at what "price" ? I admit i never tested it.
guruboolez
As promised, the bitrate table for each VBR -V mode and for both "default" & "new" VBR algorithms.
It's based on 150 classical tracks. Bitrate bloat due to sb21 are not really common; as a consequence, the average bitrate I obtained for the highest profiles is probably lower than what people encoding dynamically compressed music would get.

VBR [OLD]

CODE

V0 V1 V2 V3 V4 V5 V6 V7 V8 V9

239,0 229,8 207,1 195,2 175,8 134,7 112,6 92,4 74,5 52,5
253,4 233,9 206,0 169,4 153,4 133,4 119,4 99,7 83,2 57,8
269,3 252,5 225,0 196,8 176,8 150,4 137,5 122,6 97,6 69,4
262,4 243,8 215,4 177,1 161,4 142,9 130,7 112,5 93,2 68,4
218,9 201,3 177,9 156,4 140,4 108,3 93,5 79,6 67,1 46,4
174,0 166,8 145,5 133,9 115,3 82,1 69,0 55,1 47,2 31,0
224,9 213,3 191,8 182,7 164,9 129,8 109,2 89,8 73,7 52,7
222,0 211,1 190,4 180,2 164,1 131,8 105,7 81,2 63,9 42,9
216,4 205,3 184,0 174,3 153,4 111,7 94,4 76,7 61,7 40,8
228,4 211,4 187,0 158,1 142,2 119,2 95,5 71,2 52,4 31,9
230,5 220,4 199,0 188,4 167,6 130,5 107,8 87,7 72,0 50,6
220,6 211,8 191,1 177,3 153,8 109,5 90,3 71,1 57,7 38,1
244,9 222,2 196,9 180,9 161,4 127,2 109,4 94,2 81,8 61,9
224,5 213,0 192,3 182,9 163,7 127,6 106,5 86,4 70,4 50,2
220,3 200,3 176,7 165,0 146,7 116,2 98,1 80,1 65,2 45,9
244,4 228,8 204,5 182,5 166,5 139,7 120,2 98,7 79,3 49,8
212,4 201,5 179,6 167,7 148,4 110,6 91,9 73,3 58,1 38,1
216,5 201,8 179,9 170,5 154,1 126,6 106,1 85,1 68,9 48,1
225,9 216,7 195,9 185,9 166,8 131,7 110,4 89,7 72,6 50,6
226,8 217,7 195,9 186,2 168,9 131,5 107,7 85,2 68,9 49,0
232,3 221,3 199,2 187,0 170,6 140,2 118,6 98,3 80,7 57,2
239,2 224,7 201,7 184,6 167,1 137,5 114,9 93,0 73,8 50,6
218,7 209,0 186,8 173,4 154,3 117,3 98,8 83,0 68,9 50,0
209,6 198,2 175,1 161,3 140,1 104,5 88,3 71,3 59,1 40,6
212,4 204,4 184,5 171,3 145,1 91,3 72,9 56,1 42,9 23,3
221,2 211,5 190,0 178,7 158,0 115,1 93,9 75,4 61,3 42,1
230,5 217,4 195,5 187,4 172,5 143,8 119,5 97,3 78,4 55,0
226,5 214,9 193,2 179,2 156,6 117,5 98,8 78,6 64,8 44,0
218,6 208,5 186,5 172,7 151,8 111,9 92,5 72,3 57,7 38,8
238,6 225,2 200,9 184,4 167,5 140,8 122,7 103,5 86,9 64,2
237,2 220,0 195,6 181,7 160,4 124,2 103,2 84,3 69,9 50,7
230,0 215,9 193,0 179,2 157,1 117,2 97,9 77,9 63,4 43,4
242,2 227,7 202,7 187,6 169,1 134,4 115,0 97,8 81,6 60,2
232,2 209,3 182,7 164,7 145,8 113,3 96,2 78,3 65,9 45,5
238,8 226,7 204,3 193,3 176,3 141,9 117,4 95,6 76,4 52,9
227,2 213,1 189,2 163,9 150,0 132,2 122,4 106,0 89,5 66,1
228,5 216,7 194,6 184,5 168,2 140,9 121,2 100,7 85,7 63,6
228,1 208,3 184,1 173,7 155,9 128,3 112,6 93,3 81,8 61,0
211,6 202,9 182,0 170,5 153,8 118,9 96,0 76,9 62,2 39,4
214,4 204,4 183,1 172,4 156,0 128,8 114,2 93,6 79,2 53,8
225,9 212,3 188,9 174,7 160,2 140,3 126,9 109,5 92,2 69,5
216,4 207,1 185,6 173,6 154,7 120,5 102,2 81,4 67,6 48,1
220,6 210,9 190,2 180,0 160,4 123,7 103,1 82,5 66,8 46,0
224,8 213,5 191,4 179,1 157,9 118,2 97,9 78,5 63,5 42,6
224,7 208,0 184,5 170,9 150,9 116,9 98,3 80,0 65,9 46,3
209,2 199,5 180,0 173,1 157,6 135,1 113,8 91,9 77,8 56,3
207,9 198,5 179,4 169,4 153,5 126,8 109,7 89,9 76,0 54,6
228,9 216,7 194,0 181,9 164,2 135,2 116,1 96,3 80,9 58,1
222,3 204,5 180,5 167,4 150,1 108,8 84,8 65,0 50,4 28,7
237,3 220,7 195,0 172,5 157,1 138,5 126,2 107,2 90,7 64,8
226,2 214,7 192,3 179,3 161,0 127,4 110,5 91,2 76,5 55,4
229,7 218,8 196,9 187,2 169,5 137,8 117,5 98,6 82,6 61,6
228,8 218,1 196,5 186,7 169,0 138,1 117,9 97,3 80,7 57,9
229,2 215,1 192,1 183,4 166,6 134,8 115,8 97,8 81,4 59,7
259,4 240,7 211,0 184,6 169,3 148,2 135,5 117,6 99,2 73,7
229,4 217,8 196,1 187,9 171,0 142,3 121,7 102,5 85,6 61,3
228,5 215,1 191,8 181,9 166,5 144,5 126,6 107,6 91,4 66,4
236,5 218,3 192,3 176,7 160,3 137,6 121,0 102,5 86,9 63,0
233,3 214,6 190,2 179,2 163,0 139,3 120,4 102,0 86,1 64,4
232,1 220,9 198,9 189,9 173,3 141,7 120,0 99,4 81,4 58,6
233,6 213,8 187,6 169,6 154,1 132,2 117,1 98,5 81,7 57,4
239,5 221,2 195,1 170,5 154,4 131,0 115,7 98,9 82,9 59,2
242,9 229,8 205,0 187,1 171,2 148,2 131,9 114,6 98,6 71,8
220,5 209,1 187,5 178,7 162,9 140,2 124,8 105,7 89,3 63,2
241,3 223,1 196,1 179,0 164,2 142,9 126,3 108,7 92,6 67,2
200,8 170,6 143,3 131,0 116,6 95,2 84,6 66,0 53,5 35,8
219,2 204,1 181,8 170,9 151,2 113,0 94,2 76,6 61,3 41,0
219,5 210,6 189,9 177,9 156,5 116,6 96,8 76,0 60,0 39,6
223,8 208,9 186,3 175,7 156,1 122,8 106,2 85,6 70,9 50,8
212,3 201,1 180,2 171,7 154,9 125,1 105,0 83,2 65,8 43,6
203,1 193,1 173,0 163,2 143,7 104,4 87,3 68,5 56,2 37,7
216,5 206,3 186,0 179,0 163,4 140,0 120,1 97,1 79,4 57,4
226,0 211,3 188,3 177,0 157,5 123,1 102,8 83,1 67,5 48,2
217,4 204,0 182,9 173,9 155,6 116,7 95,7 73,7 58,5 38,4
249,4 218,6 184,9 149,5 133,7 113,1 98,0 79,0 63,6 43,7
225,8 211,6 185,6 161,6 145,7 124,3 110,8 90,4 79,6 57,3
223,6 209,2 183,1 159,4 143,6 125,4 116,3 96,7 83,7 61,3
245,2 226,4 198,2 161,9 146,4 128,3 117,7 97,7 81,3 58,3
255,2 240,0 212,8 168,6 153,8 135,0 125,8 107,5 89,3 65,4
226,4 215,6 192,4 178,5 161,7 136,7 122,3 99,6 81,2 55,1
208,8 197,4 175,0 158,6 144,1 124,0 112,9 91,6 79,3 58,5
237,3 222,0 198,7 189,8 172,9 142,4 122,3 102,0 83,9 60,1
215,6 204,8 184,8 174,1 152,3 117,7 94,8 75,2 60,2 35,1
228,6 218,6 195,8 181,7 161,3 122,2 102,6 80,9 64,5 44,4
260,0 248,1 222,3 206,2 187,0 151,9 123,4 99,7 80,2 57,9
211,1 197,9 177,1 166,7 148,3 117,4 100,9 78,9 63,7 37,7
219,3 208,5 187,3 177,7 160,4 131,0 112,1 91,2 74,6 49,1
237,5 227,4 205,1 193,3 173,2 131,4 107,9 88,1 71,2 50,3
228,2 214,5 189,6 179,7 161,3 113,5 96,8 79,9 67,5 49,9
233,2 211,7 187,0 174,4 155,1 118,1 89,1 66,2 52,7 35,8
194,3 186,9 166,2 154,7 136,5 98,2 82,3 65,2 54,4 37,8
205,3 196,6 174,4 162,0 141,8 96,8 79,8 63,8 53,7 37,1
200,7 192,0 171,0 160,3 145,4 108,2 89,4 71,9 60,1 44,0
221,0 212,1 189,5 174,1 152,5 108,3 87,9 69,4 56,3 38,3
178,9 169,8 149,3 137,6 117,5 81,1 68,7 53,7 45,9 29,1
178,9 169,8 149,3 137,6 117,5 81,1 68,7 53,7 45,9 29,1
208,2 192,8 170,7 158,7 140,9 112,1 96,0 77,9 64,8 45,5
221,4 203,4 180,1 167,4 148,5 112,0 93,6 75,3 59,9 39,2
224,3 214,9 193,9 186,2 169,5 139,1 119,7 96,9 77,5 52,3
229,3 210,5 187,1 178,3 161,7 141,7 127,5 98,3 88,5 68,9
216,1 207,9 187,2 177,3 161,0 123,8 105,6 86,6 70,7 47,8
224,8 206,5 184,1 174,8 158,0 126,1 103,8 83,7 67,7 47,3
241,7 226,2 205,2 195,3 177,6 143,0 116,0 92,6 72,3 52,0
227,8 215,7 192,1 176,9 155,3 111,3 88,8 70,6 57,2 39,4
217,9 207,5 187,0 179,1 164,3 142,6 123,7 100,0 75,3 50,0
220,4 203,4 180,0 167,2 150,4 121,8 95,5 72,5 54,8 34,0
221,3 201,9 178,2 166,5 145,9 111,8 93,0 72,7 57,1 36,3
224,8 211,5 191,2 182,0 161,5 127,1 105,9 85,8 68,5 45,9
193,9 182,1 161,5 155,1 140,6 120,3 105,1 87,6 76,6 55,0
196,4 183,8 162,1 150,5 133,0 94,9 76,2 57,0 44,1 26,2
209,6 198,1 176,9 166,8 148,0 107,6 84,6 64,5 50,2 32,3
215,1 203,7 181,9 170,3 152,5 112,2 88,7 69,3 53,4 33,3
226,6 216,1 193,7 182,5 164,0 127,2 106,0 85,9 69,3 48,9
215,2 202,9 180,4 168,3 148,4 111,3 92,4 72,5 56,4 35,7
218,5 206,7 184,9 175,7 159,3 135,9 120,3 99,3 83,8 57,3
218,6 205,8 184,7 175,4 159,1 131,3 111,5 87,0 66,0 40,7
222,8 210,6 188,9 179,5 163,2 135,7 116,4 96,2 80,2 57,5
184,8 169,7 147,4 135,3 117,7 89,0 75,5 57,8 48,5 32,3
186,9 173,3 150,5 135,9 114,0 75,9 61,9 48,6 40,2 25,0
194,8 184,4 161,3 148,9 130,9 97,5 81,5 63,2 52,0 35,7
227,7 217,7 197,0 187,4 169,6 138,3 116,8 96,2 79,2 54,3
243,6 231,4 208,1 197,0 178,6 142,0 119,8 98,8 79,7 56,5
172,9 166,7 145,0 134,5 113,0 71,5 56,5 41,3 32,9 18,9
249,3 231,4 202,9 176,0 160,4 137,5 126,3 110,6 93,5 69,5
230,7 220,2 198,6 188,4 170,4 135,7 113,7 93,2 74,8 52,2
233,1 221,5 199,7 192,2 177,2 153,0 131,9 113,4 95,5 69,4
247,2 232,5 209,3 198,2 181,9 142,4 119,7 97,6 77,3 54,8
217,4 205,5 183,7 170,2 147,5 101,7 83,3 65,4 52,0 32,5
223,0 204,1 179,6 169,5 153,3 129,0 111,5 91,0 75,1 53,2
221,0 208,1 185,5 176,1 159,7 135,3 114,6 93,7 78,5 55,1
231,8 214,3 190,3 176,3 155,8 119,7 100,4 81,1 66,0 45,7
232,3 221,8 200,3 189,6 171,5 136,6 115,9 92,0 76,0 53,4
235,9 222,0 199,4 188,4 171,9 143,3 123,3 103,6 86,3 62,0
246,5 235,1 211,7 200,4 183,0 146,3 125,0 103,8 84,0 61,7
241,4 228,6 202,0 190,5 170,3 134,3 114,5 95,6 78,8 52,1
231,5 219,9 197,4 183,7 161,2 123,0 103,1 82,3 68,3 47,8
230,8 216,6 193,7 179,2 159,1 109,8 79,8 61,0 48,4 29,7
225,8 215,7 194,7 184,0 166,3 136,2 116,5 95,9 79,4 55,0
226,4 207,7 184,2 173,2 158,0 124,6 102,2 79,7 62,8 42,7
230,7 216,7 193,7 180,5 163,0 131,1 112,0 93,9 77,1 52,9
231,5 209,8 183,5 163,7 148,5 125,6 110,9 93,1 78,3 55,7
219,3 207,8 185,6 176,0 158,5 123,0 100,8 81,4 64,6 42,7
226,2 212,8 190,4 178,7 158,3 121,4 101,4 78,6 63,5 43,8
237,4 220,3 194,2 164,9 148,8 129,9 119,8 101,5 86,3 61,7
222,2 207,8 185,8 173,6 158,7 130,2 104,8 77,8 56,8 34,9
218,8 201,9 179,3 167,6 149,9 113,7 96,2 76,1 62,8 43,2
234,2 216,1 190,7 176,0 159,1 133,4 115,9 97,0 82,0 59,7
228,2 214,6 191,2 177,4 156,1 116,5 96,4 76,9 61,1 40,7
229,7 217,1 194,5 185,1 168,7 142,8 122,2 103,0 87,5 62,9
227,3 216,1 194,5 182,4 163,7 130,4 110,4 89,2 73,4 51,5




VBR [NEW]

CODE

V0 V1 V2 V3 V4 V5 V6 V7 V8 V9

234,3 221,8 194,6 181,7 167,0 129,5 109,0 85,3 69,5 48,6
253,6 240,6 210,8 171,6 160,0 140,5 125,2 103,4 86,0 56,4
266,8 258,7 230,3 200,7 184,4 157,5 143,4 125,7 99,4 67,7
261,3 249,5 217,7 178,0 166,7 148,3 134,6 112,8 94,3 69,2
216,5 202,1 175,7 153,5 140,5 108,6 93,1 77,4 65,9 44,1
190,0 165,1 135,1 119,5 106,2 78,3 66,0 51,8 45,0 26,2
218,3 207,4 181,9 173,1 159,6 127,1 107,4 85,0 70,1 49,4
220,2 211,6 186,9 178,1 166,2 133,2 106,3 76,4 60,6 37,9
218,9 204,1 178,0 167,4 150,9 111,1 93,4 72,6 57,9 36,5
229,2 216,9 188,6 158,9 147,3 123,6 98,4 67,8 49,4 25,6
221,6 213,5 188,7 177,8 161,5 127,8 106,0 83,0 68,5 47,2
215,7 206,1 180,7 165,4 147,2 107,2 89,1 66,9 54,1 33,2
245,4 220,7 191,4 175,1 159,8 127,6 109,9 93,1 81,2 60,1
214,4 205,6 181,2 171,4 156,9 123,9 104,4 81,2 66,6 46,8
213,5 195,3 167,6 156,1 141,9 113,3 96,0 74,9 60,7 42,3
242,8 228,9 201,5 182,8 171,2 144,6 123,7 98,3 76,3 43,2
212,2 196,7 169,5 156,8 142,0 108,2 90,6 68,2 53,9 33,8
216,8 198,5 172,6 164,4 152,8 125,0 104,5 79,9 64,7 44,6
219,0 209,8 184,9 174,8 160,5 128,3 107,9 83,1 67,3 47,2
220,3 206,8 180,3 169,8 157,4 124,9 104,0 77,5 64,1 45,9
225,1 219,0 193,7 182,0 169,8 140,1 118,8 94,7 78,0 54,3
231,5 222,3 195,9 179,2 166,0 137,6 115,2 88,3 70,0 46,1
218,7 201,8 173,7 159,1 145,1 113,4 97,3 78,9 65,7 45,7
214,3 195,8 167,0 150,7 134,2 102,3 87,0 68,0 56,8 36,7
211,9 203,4 178,5 163,6 142,1 91,5 72,4 52,7 38,1 18,3
217,8 207,0 181,3 169,4 152,6 112,6 93,1 71,5 58,6 38,2
225,7 213,2 187,9 181,6 171,1 142,0 117,4 90,8 73,9 52,2
218,7 207,6 182,2 166,3 149,5 115,1 97,3 75,0 61,4 39,9
216,3 202,9 175,8 161,3 145,4 109,2 90,5 67,1 53,5 34,9
234,1 221,9 194,4 178,2 166,2 141,1 123,3 102,4 86,5 63,5
228,0 212,4 184,5 170,2 153,7 120,7 100,9 79,8 66,6 47,1
223,7 211,4 184,9 170,7 153,0 115,9 97,4 74,4 60,4 39,6
236,2 220,2 191,3 175,7 162,1 130,8 113,2 93,9 79,2 57,7
231,5 207,5 176,9 158,6 143,7 112,3 95,4 75,7 64,3 42,7
233,7 223,1 196,9 186,8 174,2 140,3 116,0 90,2 72,7 50,0
223,1 217,7 191,2 166,2 156,4 139,8 129,0 111,5 94,2 67,3
222,5 211,7 186,1 176,7 165,2 139,3 120,6 98,8 85,0 62,7
223,1 204,8 177,0 166,9 154,0 127,7 112,1 91,9 80,6 59,9
214,0 200,8 174,5 162,9 150,6 119,1 96,7 73,0 58,5 32,9
213,5 206,0 181,2 171,3 159,6 134,0 118,2 94,6 78,3 49,3
225,0 213,5 186,6 173,3 163,1 144,1 130,3 111,3 94,4 70,6
212,9 201,3 175,3 163,5 149,4 118,8 101,8 78,0 64,8 44,4
216,6 207,3 182,5 172,2 157,0 123,1 103,1 78,9 63,5 41,7
219,3 209,7 183,7 171,0 154,4 117,8 98,0 75,0 60,2 37,4
218,2 198,5 170,6 156,6 141,5 111,4 94,4 74,0 61,3 43,2
205,6 197,9 174,7 169,8 159,3 137,2 115,1 90,1 75,0 53,8
200,8 196,0 173,1 164,2 152,5 126,4 110,0 88,2 74,1 51,6
224,7 214,0 187,7 176,4 163,4 136,0 117,4 95,3 79,8 56,1
227,3 207,8 179,9 167,7 154,3 111,4 85,8 62,0 46,6 21,7
238,3 224,5 194,6 173,1 162,7 145,0 131,6 109,4 93,0 64,3
225,6 213,5 186,9 173,8 160,1 129,3 112,2 90,0 74,9 51,9
226,0 213,4 187,1 177,6 164,7 136,1 117,3 95,2 79,7 58,5
222,8 212,9 187,6 178,3 165,6 137,0 117,8 93,9 77,7 54,7
220,4 206,4 180,2 171,7 159,7 131,4 113,6 93,6 78,0 56,8
258,9 243,8 210,8 184,5 173,6 153,5 140,7 121,4 102,9 75,7
224,0 213,1 187,9 180,7 168,4 142,1 121,6 99,3 82,8 58,3
224,9 212,8 185,8 177,1 166,6 145,8 127,3 105,4 89,4 65,3
237,6 219,6 189,4 174,4 162,5 141,1 124,0 102,3 86,5 61,2
233,9 212,2 183,4 173,9 162,2 139,3 119,9 98,5 83,4 62,8
225,5 213,6 188,0 179,6 167,9 139,0 117,7 93,9 77,1 56,1
234,7 216,2 186,4 169,1 157,8 136,4 120,5 98,9 81,9 55,5
240,0 225,5 195,3 171,7 159,9 137,0 120,4 101,8 85,2 58,1
238,9 228,5 200,6 182,8 171,4 150,0 133,9 115,2 99,6 72,3
220,8 210,1 184,5 177,0 166,2 144,9 128,7 106,3 88,9 60,7
245,6 224,8 192,9 176,5 166,1 145,8 129,3 108,2 93,0 66,0
200,8 170,5 138,7 128,1 117,1 95,2 84,5 63,3 50,2 31,6
216,2 200,1 173,4 161,6 145,8 110,2 92,2 71,4 56,9 36,2
214,6 205,9 181,0 168,4 151,7 115,0 95,8 71,5 56,1 35,4
216,1 205,8 180,1 169,9 154,5 122,8 106,5 82,9 68,2 46,7
211,2 202,7 178,0 170,8 157,9 128,0 106,7 79,7 62,0 38,8
202,6 191,1 166,6 156,1 140,7 103,6 86,5 65,3 53,5 33,2
210,7 203,3 179,4 174,0 163,0 139,8 118,9 92,4 75,9 55,0
220,9 207,9 180,8 169,8 154,5 121,7 101,7 78,6 64,1 44,5
215,9 202,6 177,7 169,4 154,6 116,7 95,8 69,7 55,1 34,1
254,5 221,4 183,9 149,0 137,3 116,3 100,3 78,0 62,0 39,4
223,8 213,4 185,3 161,9 150,0 128,8 114,5 92,3 80,6 55,0
219,6 212,5 184,9 161,5 149,7 131,3 121,2 100,4 86,5 60,7
237,6 229,8 202,2 164,7 153,3 134,7 122,8 102,4 85,5 59,1
244,3 244,5 215,9 170,4 159,9 142,2 132,2 113,0 94,5 67,6
222,4 217,9 192,6 179,3 166,8 142,1 126,7 100,0 82,4 52,5
204,3 199,2 174,5 158,7 147,7 127,8 116,0 93,1 80,5 57,4
227,2 214,7 188,9 180,8 168,6 140,0 120,0 96,4 79,6 56,0
214,5 208,4 184,1 173,3 154,2 118,8 95,4 71,6 55,6 27,8
223,3 207,8 180,2 164,8 150,5 115,2 97,4 72,9 58,8 40,2
242,3 231,9 204,3 185,5 171,3 138,7 113,6 86,0 71,1 55,2
211,1 200,6 175,3 165,6 150,6 119,6 102,7 76,5 59,9 31,5
213,8 206,1 181,5 173,0 160,4 132,0 112,2 86,8 69,7 44,8
226,0 216,6 190,3 177,7 162,6 126,1 104,0 80,6 65,3 46,6
230,2 201,4 170,4 158,6 145,6 105,2 91,1 72,6 62,9 46,0
227,5 209,6 181,8 169,7 154,8 116,0 86,6 59,5 48,5 32,0
201,6 177,5 150,3 136,6 123,3 91,5 78,1 59,8 50,7 34,0
214,6 189,9 161,4 146,0 130,3 91,4 75,0 58,1 49,9 32,9
208,1 178,7 151,7 140,5 130,3 100,2 84,6 65,1 55,3 40,8
219,6 204,0 175,8 159,1 142,3 103,4 85,1 64,1 52,3 34,2
188,9 168,5 141,7 126,9 110,8 78,9 67,2 51,0 43,8 23,9
188,9 168,5 141,7 126,9 110,8 78,9 67,2 51,0 43,8 23,9
199,4 184,8 159,2 145,6 133,3 108,1 93,3 73,5 61,4 42,0
221,6 201,4 173,7 160,8 145,7 112,0 93,9 71,8 56,4 34,1
215,9 210,0 185,7 178,4 166,6 138,1 119,6 92,2 72,9 47,6
218,1 204,8 178,3 171,2 159,2 139,7 124,5 99,4 90,2 70,0
218,5 202,2 177,0 166,3 154,4 121,9 104,9 82,6 66,5 41,3
217,6 200,6 174,3 165,1 152,6 121,8 99,9 76,9 62,9 44,3
231,1 218,3 192,5 181,1 168,2 136,7 111,7 83,3 66,2 48,5
218,3 206,3 177,2 161,8 144,9 106,5 86,1 64,8 52,9 35,3
217,1 208,3 184,2 178,7 168,9 145,7 124,6 93,6 69,2 45,2
220,2 206,7 179,6 167,6 154,9 125,2 97,9 68,1 51,4 28,1
217,8 202,0 174,7 163,2 145,9 111,9 93,1 68,5 52,7 31,3
218,6 209,8 185,3 175,3 157,9 125,8 105,4 80,7 64,0 40,3
194,4 181,4 157,3 152,5 142,2 122,3 106,2 88,9 74,3 53,5
205,2 184,9 156,9 143,4 129,2 93,0 74,2 52,0 40,1 21,3
208,3 197,7 171,9 161,7 146,7 107,0 83,4 58,9 46,6 27,3
212,4 202,7 176,7 164,8 151,1 111,2 87,8 64,6 49,2 28,2
220,3 206,7 180,2 168,5 154,8 121,7 102,0 78,5 63,8 45,3
213,9 201,5 174,0 161,5 145,9 111,2 92,6 68,1 52,5 31,1
218,1 207,9 182,2 174,1 162,6 140,0 123,4 99,3 82,6 53,8
216,4 207,9 182,9 175,4 163,6 134,5 113,6 82,6 61,1 35,2
220,0 208,2 182,4 173,8 162,1 136,6 117,4 93,8 77,7 54,5
194,1 169,3 140,4 126,4 113,0 86,4 73,4 54,1 45,9 27,5
202,3 176,9 146,4 127,3 109,7 73,9 60,3 45,5 38,3 18,9
206,6 182,3 152,6 138,1 124,9 93,9 78,8 58,5 48,7 31,9
220,1 212,4 188,1 179,1 166,1 137,4 116,2 92,2 75,1 50,5
228,8 220,2 194,7 183,5 169,8 135,8 115,4 91,3 74,2 53,0
192,5 172,1 141,4 124,6 107,4 70,3 55,4 38,1 32,3 12,8
250,2 235,3 202,7 175,2 163,9 142,4 130,6 113,9 96,5 70,1
220,3 212,8 188,7 178,5 164,9 132,0 110,3 86,3 69,5 48,3
228,3 217,3 191,8 185,7 175,5 153,0 131,6 110,0 92,2 67,0
236,0 225,4 199,8 189,7 177,6 137,1 115,7 90,1 71,9 51,5
220,2 204,6 178,6 163,2 144,3 101,0 82,8 61,7 48,8 27,0
219,7 201,5 173,3 164,7 153,0 128,7 111,2 87,3 71,4 49,2
219,7 207,0 180,7 172,9 160,9 137,3 115,9 91,8 75,7 51,9
222,7 206,4 178,6 164,0 147,9 115,0 97,2 75,0 60,9 41,8
224,7 216,7 191,8 180,9 167,3 134,1 114,2 87,7 73,1 49,0
228,3 215,6 189,2 178,7 166,6 139,3 119,6 97,7 82,1 60,1
232,2 223,7 198,4 187,0 174,3 139,0 119,6 94,4 78,4 59,4
236,7 219,7 189,2 176,6 162,2 129,6 111,5 87,9 71,7 47,4
226,0 215,5 189,2 172,8 154,8 120,6 102,6 79,6 65,2 43,5
225,4 214,0 187,0 172,4 157,1 109,3 78,6 55,7 44,5 24,3
222,0 213,0 188,0 177,9 165,0 137,3 117,7 93,4 76,7 51,0
225,3 207,0 180,0 170,5 159,1 125,3 102,6 75,3 59,2 38,3
229,1 217,6 190,8 178,0 164,2 132,6 113,1 91,7 74,7 48,9
227,5 207,8 178,0 158,2 147,0 125,2 109,5 90,7 76,5 54,1
216,1 206,7 180,4 171,0 157,7 123,4 101,0 77,3 60,5 37,7
218,3 207,6 181,4 169,0 153,0 118,9 100,0 73,7 59,3 40,0
236,0 223,9 195,8 166,2 155,0 136,3 124,8 104,6 88,6 60,6
222,2 210,0 184,1 173,2 162,9 133,7 107,1 72,2 51,9 29,7
219,1 199,4 172,5 160,8 147,2 113,2 96,2 73,5 60,4 38,4
231,7 213,9 184,6 170,9 158,4 134,4 117,0 96,0 80,7 57,4
219,2 207,5 180,1 164,9 148,7 112,9 94,1 70,6 55,7 36,3
221,8 212,6 186,9 178,6 166,8 142,5 122,0 100,6 85,1 60,1
221,9 210,4 184,7 172,7 159,0 128,7 109,8 85,5 69,9 48,0


The bitrate curves look like this:

user posted image

user posted image

Each curve correspond to a -Vx mode.


To sum up:

CODE
VBR NEW
AVERAGE MIN MAX
V0 221,89 188,90 266,80
V1 208,40 165,10 258,70
V2 181,46 135,10 230,30
V3 167,69 119,50 200,70
V4 154,21 106,20 184,40
V5 124,25 70,30 157,50
V6 105,81 55,40 143,40
V7 83,17 38,10 125,70
V8 68,29 32,30 102,90
V9 46,03 12,80 75,70

CODE

VBR OLD
AVERAGE MIN MAX
V0 224,67 172,90 269,30
V1 211,18 166,70 252,50
V2 188,23 143,30 225,00
V3 174,60 131,00 206,20
V4 156,66 113,00 187,00
V5 124,79 71,50 153,00
V6 106,08 56,50 137,50
V7 80,46 41,30 122,60
V8 71,00 32,90 99,20
V9 49,45 18,90 73,10
Vietwoojagig
QUOTE(guruboolez @ Sep 20 2005, 01:37 PM)
To sum up:

CODE
VBR NEW   
      AVERAGE    MEDIAN        MIN        MAX
V0    221,89    220,30    188,90    266,80
V1    208,40    207,90    165,10    258,70
V2    181,46    182,20    135,10    230,30
V3    167,69    170,60    119,50    200,70
V4    154,21    156,95    106,20    184,40
V5    124,25    127,35      70,30    157,50
V6    105,81    106,90      55,40    143,40
V7      83,17      82,95      38,10    125,70
V8      68,29      66,95      32,30    102,90
V9      46,03      46,35      12,80      75,70

CODE

VBR OLD 
      AVERAGE    MEDIAN        MIN        MAX
V0    224,67    225,95    172,90    269,30
V1    211,18    211,95    166,70    252,50
V2    188,23    189,75    143,30    225,00
V3    174,60    176,50    131,00    206,20
V4    156,66    158,00    113,00    187,00
V5    124,79    127,20      71,50    153,00
V6    106,08    107,85      56,50    137,50
V7      80,46      87,65      41,30    122,60
V8      71,00      71,60      32,90      99,20
V9      49,45      50,25      18,90      73,10

*



I prefere the median instead of the average. It's the one in the middle and its value is more resistent against outliers.

But appart from that: great work! Now we only have to find someone who is willing to do the same thing with the other music-styles laugh.gif
guruboolez
QUOTE(Vietwoojagig @ Sep 20 2005, 03:01 PM)
I prefere the median instead of the average. It's the one in the middle and its value is more resistent against outliers.
*



Corrected smile.gif

CODE

VBR OLD || VBR NEW
AVERAGE MEDIAN MIN MAX || AVERAGE MEDIAN MIN MAX
V0 224,7 226,0 172,9 269,3 || V0 221,9 220,3 188,9 266,8
V1 211,2 211,9 166,7 252,5 || V1 208,4 207,9 165,1 258,7
V2 188,2 189,7 143,3 225,0 || V2 181,5 182,2 135,1 230,3
V3 174,6 176,5 131,0 206,2 || V3 167,7 170,6 119,5 200,7
V4 156,7 158,0 113,0 187,0 || V4 154,2 157,0 106,2 184,4
V5 124,8 127,2 71,5 153,0 || V5 124,3 127,3 70,3 157,5
V6 106,1 107,8 56,5 137,5 || V6 105,8 106,9 55,4 143,4
V7 86,5 87,6 41,3 122,6 || V7 83,2 82,9 38,1 125,7
V8 71,0 71,6 32,9 99,2 || V8 68,3 67,0 32,3 102,9
V9 49,5 50,3 18,9 73,1 || V9 46,0 46,3 12,8 75,7


EDIT: V7 "old" average bitrate: 80.5 -> 86.5 (typo error).
Klyith
QUOTE(Vietwoojagig @ Sep 20 2005, 10:01 AM)
I prefere the median instead of the average. It's the one in the middle and its value is more resistent against outliers.
Mode never gets any love. wink.gif
Also the phpbb code blocks are annoying just too small for the whole table.


Back to athaa-sensitivity: Its applications are fairly limited, and the bitrate increase is IMHO too much and too unpredictable to justify being incorporated into the presets. And yet it doesn't take up bitrate comparable to moving up one quality level. It's kind of stuck in no-man's-land.

Maybe vbr-new + athaa-sensitivity would equalize out back to the same bitrate as the old presets. But I also would be worried about regressions with it. Frankly the presets are great, but I think it is too much to expect them to be perfect for all music without increasing bitrate demands too much.

So I think it should stay as it is in the alphas, possibly with a note about use. "Setting --athaa-sensitivity to 1 can improve quality with some types of very quiet music at the cost of some bitrate increase."
guruboolez
QUOTE(Klyith @ Sep 20 2005, 04:02 PM)
Back to athaa-sensitivity: Its applications are fairly limited, and the bitrate increase is IMHO too much and too unpredictable to justify being incorporated into the presets.

And? VBR is by itself unpredicatable. Just take a look on the datas I've just posted. On the contrary, the --athaa-sensitivity 1 command is just limiting the amplitude of the variation, by preventing the encoder to use too low bitrate.

QUOTE
Maybe vbr-new + athaa-sensitivity would equalize out back to the same bitrate as the old presets. But I also would be worried about regressions with it. Frankly the presets are great


I'd suggest you to test current presets with and without --athaa-sensitivity 1... They're far to be great.
Most often, there's no difference in bitrate. But when a difference occurs, be sure that the additionnal bits are worth.

QUOTE
So I think it should stay as it is in the alphas, possibly with a note about use. "Setting --athaa-sensitivity to 1 can improve quality with some types of very quiet music at the cost of some bitrate increase."


In other word, you're suggesting that people should use an alpha encoder for safety?!
BTW, I recall that the quality gain is noticeable on quit PASSAGE, and not only with quiet tracks. As long as music still have some dynamic range, the current low/mid bitrate presets are not fully safe to use if compared to ABR/CBR.
Lyx
QUOTE
In other word, you're suggesting that people should use an alpha encoder for safety?!

I agree. Telling users to use an alpha to make use of a custom switch is the wrong solution. I also agree with the decision of the lame devs to remove --athaa-sensitivity from non-alpha builds. This is the kind of switch about which users should not need to care. If its necessary, then the encoder should automatically use it. If not, then it shouldn't. So, the question is only if and when it should be defaulted, not if the switch should be reimplemented.

The user should not need to "tune" an encoder. This is the job of the encoder. An encoder which needs additional switches for tuning is flawed and needs to be fixed. The exception obviously are developer- and tester-builds which by definition are meant to experiment.

- Lyx
AtaqueEG
QUOTE
V4  154,2  157,0  106,2  184,4
V5  124,3  127,3  70,3  157,5


This is a very dramatic difference.
Surely something could be done to bridge the gap between this two quality settings?

Klyith
QUOTE(guruboolez @ Sep 20 2005, 11:14 AM)
I'd suggest you to test current presets with and without --athaa-sensitivity 1... They're far to be great.
Most often, there's no difference in bitrate. But when a difference occurs, be sure that the additionnal bits are worth.
I will do so. But I can really only test at trying them out and seeing bitrate results. I am horrible at ABX quality tests. I never make quality opinions that would require TOS8, because I know I can't back it up.

But I have limited material to test on. My classical collection is only about 30 cds, heavily weighted to 18th-19th century. If this is something that is most evident on your harpsichord music, I give up. (I hate the harpsichord smile.gif) I'll also throw some other stuff at it to see if I can get it to trigger in other circumstances... Some electronic music might do the trick.

QUOTE(guruboolez)
In other word, you're suggesting that people should use an alpha encoder for safety?!
BTW, I recall that the quality gain is noticeable on quit PASSAGE, and not only with quiet tracks.
Not that people should stay with alpha, only that this needs to be looked at in more detail. Until a decision is made, why not keep the switch exposed to keep the people who want to use a-s 1 happy? And yes, I am aware of how the encoder works in general principles.


QUOTE(Lyx @ Sep 20 2005, 12:15 PM)
I agree. Telling users to use an alpha to make use of a custom switch is the wrong solution. I also agree with the decision of the lame devs to remove --athaa-sensitivity from non-alpha builds. This is the kind of switch about which users should not need to care. If its necessary, then the encoder should automatically use it. If not, then it shouldn't. So, the question is only if and when it should be defaulted, not if the switch should be reimplemented.
But if the decision is made not to make it default, what do the people like guru who use and like this switch do?

QUOTE(Lyx)
The user should not need to "tune" an encoder. This is the job of the encoder. An encoder which needs additional switches for tuning is flawed and needs to be fixed. The exception obviously are developer- and tester-builds which by definition are meant to experiment.

- Lyx
Then why is cbr 128 still the default encode? You have to enter at least one option to get a better quality result. But yes, I guess that too many switches exposed in the --longhelp is why we still have people who come in to tell us about their wonderful 20 parameter long encoding string. Still, there's a problem in principle that you can't have transparent quality without a lot of bits. When you need to encode to lower bitrate, something has to get left out.
Lyx
QUOTE(Klyith @ Sep 20 2005, 08:45 PM)
QUOTE(Lyx)
The user should not need to "tune" an encoder. This is the job of the encoder. An encoder which needs additional switches for tuning is flawed and needs to be fixed. The exception obviously are developer- and tester-builds which by definition are meant to experiment.

- Lyx
Then why is cbr 128 still the default encode? You have to enter at least one option to get a better quality result...

There are two misunderstandings here:

1. I wasn't implying that with an encoder the user should not need to specify any options at all. What i meant was that the user should only need to tell the encoder *what* he wants, but not "how to achieve it". The latter part is the job of the encoder. In practice, this means that in an optimal world only quality/target bitrate should be *necessary* with bitrate-upper-limit(for streaming), number of channels and samplerate being other optional parameters. Thus, the user should specify what he needs and the encoder shall come up with the best way to deliver it.

2. I am aware that lame still has lots of switches which imho shouldn't be there. Same for some defaults. But that does that mean that you may never "start somewhere" with improving the situation? It's a process, not a 0/1-switch.

Concerning your question what shall happen if --athaa-sensitivity is NOT made the default, although evidence is brought up that it should be: Then the encoder is flawed and should be fixed. Exposing a developer-switch so that the users can fix it is not the solution. In the worst case(which i highly doubt will happen) you will have an official lame release without it being the default, and then a ha.org-modified version which has it by default. This, from a psychological as well as objective point of view(the world does not just consist of ha.org - hardly any outsider will care about the minor difference) would still be better than "letting the users fix it by exposing the switch".

It plain simply comes down to this: if the defaults are wrong, its a bug. You dont fix bugs by requiring users to patch the code.
Alex B
Here's my little LAME 3.97b1 bitrate test. I gathered some rock/pop/latin/jazz/electronic tracks from my lossless archive. The selection was quite random, but I left out certain tracks that would have probably resulted extremely high or low bitrates. I ended up with 25 diverse sample tracks. I should have included more tracks, but this was about the best I could do this time.

I encoded them using V values from 0 to 9 with and without VBR New. Encoding each set of 25 tracks from the Monkey's Audio source files took about 13-14 minutes with VBR Old and 8-10 minutes with VBR New using a P4/2.8GHz. I didn't bother to measure exact times, because I used the PC for various other things during the encoding.


VBR OLD (-V 0 ... -V 9)

CODE
V0 V1 V2 V3 V4 V5 V6 V7 V8 V9
252 232 204 172 157 139 127 112 93 69
256 230 199 179 162 139 122 105 90 66
261 235 206 160 143 125 118 103 86 64
255 238 212 197 181 149 129 110 92 68
237 222 195 181 166 146 135 119 101 75
241 218 195 150 135 120 112 97 83 62
262 238 213 154 139 122 114 98 82 62
250 229 202 169 154 136 123 106 87 60
221 206 184 167 151 127 114 96 82 56
268 235 203 159 144 128 120 105 85 64
227 203 179 147 133 118 109 94 81 62
236 211 185 150 135 118 109 95 80 60
225 209 185 174 158 130 111 92 77 54
253 240 216 205 188 165 146 128 107 78
266 248 226 162 146 128 118 102 85 64
244 228 206 199 182 146 125 106 88 65
246 232 207 187 171 151 136 117 95 69
234 207 176 151 135 118 109 93 77 55
261 234 206 164 148 131 123 108 90 67
229 204 180 168 154 129 108 89 75 52
224 198 174 140 126 110 102 88 76 57
268 239 212 155 140 124 116 103 87 68
245 215 186 148 133 117 110 94 80 61
248 236 211 184 166 145 136 121 102 77
218 206 187 180 164 143 124 108 91 67

IPB Image


VBR NEW (-V 0 --vbr-new ... -V 9 --vbr-new)

CODE
V0N V1N V2N V3N V4N V5N V6N V7N V8N V9N
251 234 203 169 159 141 129 113 95 70
259 229 194 173 160 139 121 103 89 66
255 238 207 161 148 131 122 107 89 65
246 233 205 190 179 147 126 106 90 67
236 223 193 181 169 151 138 122 106 78
238 221 195 150 139 124 115 100 85 63
256 241 216 155 143 126 117 102 85 63
249 229 199 166 155 137 123 105 86 61
220 209 183 166 154 132 119 98 82 54
266 237 203 155 144 128 120 105 87 66
226 205 179 146 136 120 111 96 82 62
232 213 184 150 138 121 111 96 82 61
219 204 177 167 155 129 109 87 72 52
246 238 212 201 189 165 147 126 106 78
258 251 228 162 150 132 121 105 88 65
233 222 197 190 178 143 123 101 85 62
247 235 206 188 176 156 140 117 96 68
236 210 176 152 139 122 111 95 78 55
261 238 207 164 152 135 126 111 93 69
223 202 175 165 154 130 107 86 72 51
219 200 174 140 129 113 105 91 77 58
263 243 214 154 143 127 119 106 90 70
244 217 184 147 135 119 111 95 82 62
244 241 215 188 174 153 143 127 107 79
208 201 178 173 162 141 121 103 87 66

IPB Image


SUMMARY

CODE
OLD V0 V1 V2 V3 V4 V5 V6 V7 V8 V9
Min 218 198 174 140 126 110 102 88 75 52
Max 268 248 226 205 188 165 146 128 107 78
Average 245 224 198 168 152 132 120 104 87 64
Median 246 229 202 167 151 129 118 103 86 64
NEW V0N V1N V2N V3N V4N V5N V6N V7N V8N V9N
Min 208 200 174 140 129 113 105 86 72 51
Max 266 251 228 201 189 165 147 127 107 79
Average 241 225 196 166 154 134 121 104 88 64
Median 244 229 197 165 154 132 121 103 87 65

All together and a table with track list:

IPB Image IPB Image

Click the images to enlarge.


Edit: fixed the image links.
[proxima]
This should give a rough estimation of the average bitrate with VBR presets.
Also sfb21 bloat is very noticeable looking at the graphs, starting from -V2 some tracks are bloated (7,10,15). Very interesting wink.gif
Thanks to guruboolez and Alex B for these graphs.
askoff
QUOTE(AtaqueEG @ Sep 20 2005, 06:50 PM)
QUOTE
V4  154,2  157,0  106,2  184,4
V5  124,3  127,3  70,3  157,5


This is a very dramatic difference.
Surely something could be done to bridge the gap between this two quality settings?
*


I wouldn't say it's so dramatic. If you use CBR settings, there's also 128kbps and 160kbps and nothing between them.
Lyx
QUOTE(askoff @ Sep 21 2005, 03:00 PM)
QUOTE(AtaqueEG @ Sep 20 2005, 06:50 PM)
QUOTE
V4  154,2  157,0  106,2  184,4
V5  124,3  127,3   70,3   157,5


This is a very dramatic difference.
Surely something could be done to bridge the gap between this two quality settings?
*


I wouldn't say it's so dramatic. If you use CBR settings, there's also 128kbps and 160kbps and nothing between them.
*


IMHO the problem is not that the gap between V4 and V5, but the neglible difference in terms of average bitrate between V4 and V3. I'd say make V3 the new "medium", but with slightly higher quality than in the past, then lower the quality of V4 bit.

- Lyx
Vietwoojagig
QUOTE(Lyx @ Sep 21 2005, 02:09 PM)
QUOTE(askoff @ Sep 21 2005, 03:00 PM)
QUOTE(AtaqueEG @ Sep 20 2005, 06:50 PM)
QUOTE
V4  154,2  157,0  106,2  184,4
V5  124,3  127,3  70,3  157,5


This is a very dramatic difference.
Surely something could be done to bridge the gap between this two quality settings?
*


I wouldn't say it's so dramatic. If you use CBR settings, there's also 128kbps and 160kbps and nothing between them.
*


IMHO the problem is not that the gap between V4 and V5, but the neglible difference in terms of average bitrate between V4 and V3. I'd say make V3 the new "medium", but with slightly higher quality than in the past, then lower the quality of V4 bit.

- Lyx
*



Yes, you are right, if you make a linear regression using the median and vbr-new, you get something like bitrate=226.01-19.815*vbr-mode.

In this case, the highest difference is V4 which is 10kbits to high. But only if you want a linear relation between vbr-mode and target-bitrate.
rutra80
QUOTE(Assault @ Sep 19 2005, 01:04 PM)
Please let me repeat AndyMutz's question:

QUOTE
when using --vbr-new the -q switch produces exact same results, no matter if i use q0,q1,q2,q3 or q4. is that intended?


Does anyone know anything about this? Thanks for your help!
*


Here it produces different results. Try putting -q in the middle (after -V and before --vbr-new).

EDIT: I tried -q0 vs -q5. Values from 0 to 4 indeed seem to produce the same results, from 6 to 9 too. So it looks like effectively there are three -q levels with --vbr-new: 0-4, 5, 6-9.
AndyMutz
QUOTE(rutra80 @ Oct 3 2005, 03:18 PM)

EDIT: I tried -q0 vs -q5. Values from 0 to 4 indeed seem to produce the same results, from 6 to 9 too. So it looks like effectively there are three -q levels with --vbr-new: 0-4, 5, 6-9.

*



i would still like to know if this is intended or if it's just a bug which will be corrected in one of the next versions..

-andy-
Madrigal
QUOTE(AndyMutz @ Oct 5 2005, 07:05 AM)
i would still like to know if this is intended or if it's just a bug which will be corrected in one of the next versions..

-andy-
You might want to try posting your question again, in this thread.

Regards,
Madrigal
Gabriel
Intended
benc
I notice that MP2 decoding is still broken in 3.97b1:

CODE
D:\tmp>lame
LAME 32bits version 3.97 (beta 1, Sep 29 2005) (http://www.mp3dev.org/)

usage: lame [options] <infile> [outfile]

   <infile> and/or <outfile> can be "-", which means stdin/stdout.

Try:
    "lame --help"           for general usage information
or:
    "lame --preset help"    for information on suggested predefined settings
or:
    "lame --longhelp"
 or "lame -?"              for a complete options list

D:\tmp>lame --decode test.mp2
Error reading headers in mp3 input file test.mp2.

D:\tmp>lame --decode --mp2input test.mp2
Error reading headers in mp3 input file test.mp2.


Is this something that could be fixed for the next/final release?
john33
QUOTE(benc @ Oct 7 2005, 09:34 AM)
I notice that MP2 decoding is still broken in 3.97b1:

Is this something that could be fixed for the next/final release?
*


I've uploaded an amended .exe only including the corrected get_audio.c source to Rarewares mp3 section. smile.gif
john33
I have revisited the decoding issue and discovered that decoding has actually been broken since 3.94beta. The 3.94 beta was OK, but 3.95 and onwards is screwed.

The version I uploaded yesterday was fixed for mp2 but this broke mp3 decoding!!

I have just uploaded a new version, with amended get_audio.c source, with mp2 and mp3 decoding working correctly. smile.gif
benc
Interestingly the version of LAME I have on my linux installation works fine for MP2 decoding even though it says it's 3.96.1. I'm too much of a newbie to compile these things myself so I probably installed it from the livna RPM repository (I am using Fedora Core 3).
Gabriel
MP1/2 decoding are disabled by default because of the poor error resilience of our decoder.
Enabling mp1/2 decoding sometimes break mp3 decoding because the decoder finds mp1/2 sync words inside mp3 files.
rjamorim
Maybe you can replace your decoder with foobar's fixed mpg123?

Or maybe mpadec, which is also based in mpg123.
Gabriel
QUOTE
Maybe you can replace your decoder

Maybe we can find a brave soul to do it?

Replacing the decoder is planned, but low priority.
Yakhobian
QUOTE(AndyMutz @ Oct 5 2005, 11:05 PM)
QUOTE(rutra80 @ Oct 3 2005, 03:18 PM)

EDIT: I tried -q0 vs -q5. Values from 0 to 4 indeed seem to produce the same results, from 6 to 9 too. So it looks like effectively there are three -q levels with --vbr-new: 0-4, 5, 6-9.

*



i would still like to know if this is intended or if it's just a bug which will be corrected in one of the next versions..

-andy-
*


QUOTE(Gabriel @ Oct 5 2005, 11:33 PM)
Intended
*



Hi all, first post here so please go easy on me smile.gif
I'm curious as to know what the differences are now between the 3 settings and which is the best to choose? If -V2/--preset standard uses q3 by default and -h changes it to q2, then I would presume the top option out of the 3 should be chosen. However, which of the previous settings have q0-q4 merged to? If it has gone to q2 or q3, does this mean the extra lossless compression in q0 is no longer used? (Forgive me if I've got my facts wrong on the previous version)

Thanks guys and I look forward to being able to contribute to the community smile.gif
zan2828
I noticed some strange behavior using the new 3.97b1 compared to 3.90.3. When I encode using EAC, the bitrate setting dropdown box adds a -b switch to the -V 2 --vbr-new command line.

For example, if I select 128kbps the base bitrate of the mp3 will be 128. I have confirmed this using different bitrates in the dropdown box. I have read the faq stating the dropdown box in EAC compression options has no effect on the command line except if 320kbps is selected, however, it seems with 3.97b1 this does not apply i.e. the dropdown box does affect base bitrates.

I have tried to select "User Defined Encoder" but under that setting LAME just pops up for a split second and disappears when it tries to encode.

Granted I have only used these two versions of LAME so far so I don't know in which exact version this "bug" popped up. Or am i mistaken and this behavior is intended?
skelly831
QUOTE(zan2828 @ Oct 9 2005, 05:37 PM)
I noticed some strange behavior using the new 3.97b1 compared to 3.90.3. When I encode using EAC, the bitrate setting dropdown box  adds a -b switch to the -V 2 --vbr-new command line.

For example, if I select 128kbps the base bitrate of the mp3 will be 128. I have confirmed this using different bitrates in the dropdown box. I have read the faq stating the dropdown box in EAC compression options has no effect on the command line except if 320kbps is selected, however, it seems with 3.97b1 this does not apply i.e. the dropdown box does affect base bitrates.

I have tried to select "User Defined Encoder" but under that setting LAME just pops up for a split second and disappears when it tries to encode.

Granted I have only used these two versions of LAME so far so I don't know in which exact version this "bug" popped up. Or am i mistaken and this behavior is intended?
*



Maybe when you use "User Defined Encoder", you don't add "%s %d" to the end of your command line.
Dr. TaaDow
I was testing out 3.97b1 and I keep getting the following error on all tracks I'm trying to encode...

Command: C:\Apps\RazorLame\Lame.exe -V 2 "C:\Wavs\Johnny Fiasco Feat Dulce Caramelo - Move It (Original Mix).wav" "C:\Wavs\Johnny Fiasco Feat Dulce Caramelo - Move It (Original Mix).mp3"
LAME 3.97 (beta 1, Sep 29 2005) 32bits (http://www.mp3dev.org/)
CPU features: MMX (ASM used), 3DNow! (ASM used), SSE (ASM used), SSE2
Using polyphase lowpass filter, transition band: 18671 Hz - 19205 Hz
Encoding C:\Wavs\Johnny Fiasco Feat Dulce Caramelo - Move It (Original Mix).wav
to C:\Wavs\Johnny Fiasco Feat Dulce Caramelo - Move It (Original Mix).mp3
Encoding as 44.1 kHz VBR(q=2) j-stereo MPEG-1 Layer III (ca. 7.3x) qval=3
RazorLame encountered an unknown message from LAME while trying to encode "C:\Wavs\Johnny Fiasco Feat Dulce Caramelo - Move It (Original Mix).wav"!

Encoded 0 files in 0:00:52
There was an unexpected LAME message for one file, please check log for error messages.

I've also tried using -V 2 --vbr-new and it give the same error

has anyone else had this problem with 3.97b1 ?
Lyx
QUOTE(Dr. TaaDow @ Oct 10 2005, 03:09 AM)
I was testing out 3.97b1 and I keep getting the following error on all tracks I'm trying to encode...

Try to do a search and you will find a thread explaining the "error".

Short version: there was no error. It's just razorlame having crappy "problem-detecion": it just treats everything which it doesn't know yet as a problem.
Kuriboid
QUOTE(zan2828 @ Oct 9 2005, 06:37 PM)
I noticed some strange behavior using the new 3.97b1 compared to 3.90.3. When I encode using EAC, the bitrate setting dropdown box  adds a -b switch to the -V 2 --vbr-new command line.

For example, if I select 128kbps the base bitrate of the mp3 will be 128. I have confirmed this using different bitrates in the dropdown box. I have read the faq stating the dropdown box in EAC compression options has no effect on the command line except if 320kbps is selected, however, it seems with 3.97b1 this does not apply i.e. the dropdown box does affect base bitrates.

I have tried to select "User Defined Encoder" but under that setting LAME just pops up for a split second and disappears when it tries to encode.

Granted I have only used these two versions of LAME so far so I don't know in which exact version this "bug" popped up. Or am i mistaken and this behavior is intended?
*


It's just something EAC does for the "LAME MP3 Encoder" setting. Switch to "User Defined Encoder" and add %s %d like skelly831 suggested.

QUOTE(Lyx @ Oct 9 2005, 07:14 PM)
QUOTE(Dr. TaaDow @ Oct 10 2005, 03:09 AM)
I was testing out 3.97b1 and I keep getting the following error on all tracks I'm trying to encode...

Try to do a search and you will find a thread explaining the "error".

Short version: there was no error. It's just razorlame having crappy "problem-detecion": it just treats everything which it doesn't know yet as a problem.
*


Here's the RazorLame fix: http://www.dors.de/phpbb/viewtopic.php?p=1743#1743
Dr. TaaDow
the fix worked... thanks for the info
john33
QUOTE(john33 @ Oct 8 2005, 12:54 PM)
I have revisited the decoding issue and discovered that decoding has actually been broken since 3.94beta. The 3.94 beta was OK, but 3.95 and onwards is screwed.

The version I uploaded yesterday was fixed for mp2 but this broke mp3 decoding!!

I have just uploaded a new version, with amended get_audio.c source, with mp2 and mp3 decoding working correctly. smile.gif
*


I have taken yet a further look at the decoding and have now uploaded another version that now outputs wave files from decoded mp2 and mp3 files that contain the same number of samples as the outputs from foobar2000. Error checking, etc., remains as before. The d/l includes the amended get_audio.c and main.c sources.
Madrigal
QUOTE(Kuriboid @ Oct 9 2005, 08:24 PM)
Here's the RazorLame fix:
An even better "fix" is to abandon RazorLame in favor of Speek's ALL2LAME, which never has this problem, and doesn't need to be updated with each new version of LAME.

Regards,
Madrigal
john33
QUOTE(john33 @ Oct 10 2005, 01:47 PM)
I have taken yet a further look at the decoding and have now uploaded another version that now outputs wave files from decoded mp2 and mp3 files that contain the same number of samples as the outputs from foobar2000. Error checking, etc., remains as before. The d/l includes the amended get_audio.c and main.c sources.
*


And another version is now at Rarewares. I believe this version correctly handles 'enc_padding', decoding to the correct number of output wave samples from mp3 input. The outputs from decoding mp2 and mp3 files are exactly the same size as from foobar2000.

Amended source files are included.
xmixahlx
why not just drop decoding in lame until you have a decoder worth including?
Gabriel
Because we need it for the frame analyser.
john33
QUOTE(john33 @ Oct 12 2005, 11:19 AM)
And another version is now at Rarewares. I believe this version correctly handles 'enc_padding', decoding to the correct number of output wave samples from mp3 input. The outputs from decoding mp2 and mp3 files are exactly the same size as from foobar2000.

Amended source files are included.
*


What I think will be the final version of this is now at Rarewares. Includes exe, amended source files and diff files for the amendments.
Gabriel
QUOTE
What I think will be the final version of this is now at Rarewares. Includes exe, amended source files and diff files for the amendments.

I have the feeling that you are assuming that the mp2 delay is the one of current TooLame version and the mp3 delay the one of current Lame version.
Regarding mp2, this might cut the beginning of files produced with another encoder, and regarding mp3 the situation is the same unless you are reading the encoder delay field of the header.
This is a "lo-fi" version of our main content. To view the full version with more information, formatting and images, please click here.
Invision Power Board © 2001-2008 Invision Power Services, Inc.