Small piece of Fighter Beat track, looped several times. It kills all lossy codecs on all beatrates.
It's not enough to say it, you need to provide ABX logs in order to comply with our terms, especially when making incredible claims like the one you just made.
BTW what's a beatrate?
There was a thread where this track was mentioned: http://www.hydrogenaudio.org/forums/index....showtopic=57783 (http://www.hydrogenaudio.org/forums/index.php?showtopic=57783) , posts 10-14
There was a thread where this track was mentioned: http://www.hydrogenaudio.org/forums/index....showtopic=57783 (http://www.hydrogenaudio.org/forums/index....showtopic=57783) , posts 10-14
Yes, I know. I just looped the most hard for encoders part of it, so that the difference can be heard more clearly.
It's not enough to say it, you need to provide ABX logs in order to comply with our terms, especially when making incredible claims like the one you just made.
My iRiver player does not support lossless, so I use Vorbis q10, and I can hear the difference between it and Vorbis Q9, Musepack q10 and Nero AAC Q1 (max. bitrate). It looks like converted tracks sounds slightly quieter, almost indistinguishable. Dont know why, but on PC it is harder to hear this difference (maybe because it is AC97?). Probably, I should try test it more strenuously.
Anyway, this sample is the best I could find for testing audio codecs for myself, and I hope people who make listening tests pay attention to it.
BTW what's a beatrate?
LOL, the same as Fighter Bit.
I quickly tested(Not ABX) the sample with Vorbis at 160Kbps and Opus at 160Kbps CVBR.
The Vorbis encode sounds very different from the original. With the Opus encode I don't know if I am actually hearing a difference or not. I probably will have to do an ABX.
I quickly tried both Nero from 7-Feb 2006 using -q 0.3 -lc and Lame 3.98.4 -V3 and do not believe I would be able to pass an ABX test.
I quickly tried both Nero from 7-Feb 2006 using -q 0.3 -lc and Lame 3.98.4 -V3 and do not believe I would be able to pass an ABX test.
Same here; files I produced with these parameters all sound alike to me:
Oggenc 2.87/libVorbis 1.3.3 -q 0.5 [same as -b 160]
NeroAAC 1.5.1.0 -q 0.57
LAME 3.99.5 -b 320
Fraunhofer MP3S 1.5/lib4.01.01 -br 320000 -q 1
So yeah, "kills all lossy codecs on all b
trates" is an overstatement. I have to question the reports in the aforementioned thread as well.
@softrunner - Please take the time to run ABX tests in foobar2000. In the meantime, I suggest this thread be retitled by an admin to just "Fighter Beat loop" with the subtitle "possible problem sample for Ogg Vorbis encoder; ABX testing needed". Or something along those lines.
5/5 first time with both LAME -V4 and Vorbis -q5, both the most recent rarewares releases. I actually got LAME backwards, 0/5, but I could tell the difference. The encode sounded marginally better than the original to me... >_>
Never thought that such a regular sample could be that badass. Encoding with lame -V0 gives a bitrate of 314, so it certainly knows it is up against something. I've never seen anything like that.
Some of my highest-bitrate FLACs, do end up between 245 and 255 when lame -0'ed. Including the least-compressible harpsichord piece I have.
In the lossless department: TAK -p4m encodes at 1037. FLAC -8 encodes at 978, that's 5.7% better. WavPack extra high at 837. That's a 14.5 percent better than FLAC -8 and 19 percent better than TAK -p4m. I have never before seen any sample where TAK -p4m is beaten by 200 kb/s.
Ok, now I've done an ABX test of MP3 on PC, because MP3 is too obviouse to recognize.
Codec is LAME3.99. Original file was converted into MP3 320 kbps (maximum quality) with Foobar2000:
(http://s12.postimage.org/5dnfqcpq1/111.jpg) (http://postimage.org/image/5dnfqcpq1/)
foo_abx 1.3.4 report
foobar2000 v1.1.13
2012/07/21 01:31:58
File A: S:\320 GB HDD\Аудио-тесты\NEW\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Fighter_Beat_Loop.mp3
01:31:58 : Test started.
01:37:11 : 01/01 50.0%
01:39:35 : 02/02 25.0%
01:55:14 : 03/03 12.5%
01:57:05 : 04/04 6.3%
01:59:20 : 05/05 3.1%
02:01:50 : 06/06 1.6%
02:04:59 : 07/07 0.8%
02:08:29 : 08/08 0.4%
02:09:40 : 09/09 0.2%
02:13:03 : 10/10 0.1%
02:16:46 : 11/11 0.0%
02:18:15 : 12/12 0.0%
02:19:17 : 13/13 0.0%
02:20:25 : 14/14 0.0%
02:23:11 : 15/15 0.0%
02:23:46 : 16/16 0.0%
02:25:42 : Test finished.
----------
Total: 16/16 (0.0%)
(http://s16.postimage.org/uoqrqdimp/image.jpg) (http://postimage.org/image/uoqrqdimp/)
The difference in files is that in the original sample there is a sharp background scratch, and in the mp3 sample it is smoothed so that I just cannot here this information there.
Here is another test:
Vorbis Q7 (333 kbps)
(http://s13.postimage.org/nqfo6pqub/q7_file.jpg) (http://postimage.org/image/nqfo6pqub/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/22 20:38:10
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\Fighter_Beat_Loop (Vorbis q7).ogg
20:38:10 : Test started.
20:38:57 : 00/01 100.0%
20:39:15 : 01/02 75.0%
20:39:57 : 02/03 50.0%
20:40:07 : 03/04 31.3%
20:41:05 : 04/05 18.8%
20:41:44 : 05/06 10.9%
20:42:25 : 06/07 6.3%
20:42:45 : 07/08 3.5%
20:43:44 : 08/09 2.0%
20:45:22 : 09/10 1.1%
20:46:22 : 10/11 0.6%
20:48:02 : 11/12 0.3%
20:48:55 : Test finished.
----------
Total: 11/12 (0.3%)
The same story: no sharpness in the background noise (right channel).
Have you tried Opus with this sample?
Vorbis q8 (bitrate 376 kbps):
(http://s18.postimage.org/gjn62bfat/file.jpg) (http://postimage.org/image/gjn62bfat/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/22 21:01:15
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\Fighter_Beat_Loop (Vorbis q8).ogg
21:01:15 : Test started.
21:02:51 : 01/01 50.0%
21:03:28 : 02/02 25.0%
21:05:15 : 03/03 12.5%
21:05:52 : 04/04 6.3%
21:06:21 : 05/05 3.1%
21:07:00 : 06/06 1.6%
21:08:19 : 07/07 0.8%
21:08:46 : 08/08 0.4%
21:09:16 : 09/09 0.2%
21:10:01 : 10/10 0.1%
21:10:24 : Test finished.
----------
Total: 10/10 (0.1%)
Have you tried Opus with this sample?
I'll do it later.
Provide results for mpc and aac and I think we've got our bases covered.
Musepack q8 (350 kbps)
(http://s12.postimage.org/re5mcim5l/file.jpg) (http://postimage.org/image/re5mcim5l/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/23 03:16:51
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\Fighter_Beat_Loop (Musepack q8).mpc
03:16:51 : Test started.
03:17:42 : 01/01 50.0%
03:18:15 : 02/02 25.0%
03:18:29 : 03/03 12.5%
03:19:18 : 04/04 6.3%
03:19:48 : 05/05 3.1%
03:20:04 : 06/06 1.6%
03:20:40 : 07/07 0.8%
03:21:03 : 08/08 0.4%
03:21:34 : 09/09 0.2%
03:22:03 : 10/10 0.1%
03:22:13 : Test finished.
----------
Total: 10/10 (0.1%)
Nero AAC q0.9 (374 kbps)
(http://s11.postimage.org/6t466eubz/file.jpg) (http://postimage.org/image/6t466eubz/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/23 03:36:14
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\Fighter_Beat_Loop (Nero AAC q0.9).m4a
03:36:14 : Test started.
03:37:01 : 01/01 50.0%
03:38:30 : 02/02 25.0%
03:40:04 : 03/03 12.5%
03:41:25 : 04/04 6.3%
03:42:28 : 05/05 3.1%
03:44:08 : 06/06 1.6%
03:45:40 : 07/07 0.8%
03:47:36 : 07/08 3.5%
03:48:16 : 07/09 9.0%
03:48:52 : 08/10 5.5%
03:49:19 : 09/11 3.3%
03:50:36 : 10/12 1.9%
03:52:06 : 11/13 1.1%
03:53:02 : 12/14 0.6%
03:53:29 : Test finished.
----------
Total: 12/14 (0.6%)
It looks like the amplitude of a saw-like sound in aac sample is lower than in an original one.
It looks like the amplitude of a saw-like sound in aac sample is lower than in an original one.
Please reassure me that I'm reading this wrong, or somehow out of context.
I couldn't ABX lame -v2. But its a hard sample (for me at least) because its so noise-like.
Looking at it in the time domain is very interesting though. With the lowpass disabled, the time domain versions of the MP3 and lossless look nearly identical, with just a tiny bit of extra quantization noise in the lossy version. I would have expected the impulses to give the transform more trouble, but that doesn't seem to be the case.
I can't do it with AAC at v-5.
Please reassure me that I'm reading this wrong, or somehow out of context.
This is just the turn of speech. What I meant is that the wavy signal, that sounds like a power-saw, in AAC has less max. volume than in the original sample.
Opus 192 kbps (289 kbps in real) (libopus 0.9.11-119-g1a50ad0-exp_analysis)
(Converted via "opusenc.exe --music --bitrate 192")
(http://s13.postimage.org/vuo77728z/file.jpg) (http://postimage.org/image/vuo77728z/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/23 15:42:38
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\Fighter_Beat_Loop (Opus 192 kbps).opus
15:42:38 : Test started.
15:43:31 : 01/01 50.0%
15:44:57 : 01/02 75.0%
15:45:39 : 02/03 50.0%
15:46:22 : 03/04 31.3%
15:48:02 : 04/05 18.8%
15:48:25 : 05/06 10.9%
15:48:51 : 06/07 6.3%
15:50:28 : 06/08 14.5%
15:50:56 : 07/09 9.0%
15:51:55 : 08/10 5.5%
15:53:07 : 08/11 11.3%
15:54:06 : 09/12 7.3%
15:56:30 : 10/13 4.6%
15:58:24 : 11/14 2.9%
15:59:46 : 12/15 1.8%
16:03:34 : 13/16 1.1%
16:03:39 : Test finished.
----------
Total: 13/16 (1.1%)
I couldn't ABX lame -v2.
I can't do it with AAC at v-5.
You can focus your eyes on some object, but the most information still comes from peripheral seeing. The same in hearing. Dont try to focus on particular sounds, otherwise you will miss the most information. That's how I do it: just trying to listen via unfocused peripheral hearing, and the sound shows its difference itself.
Now I'm getting curious: can you nail down a freeformat@640?
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/23 15:42:38
File A: S:\320 GB HDD\?????-?????\NEW\Converter\input\Fighter_Beat_Loop.wav
File B: S:\320 GB HDD\?????-?????\NEW\Converter\output\Fighter_Beat_Loop (Opus 192 kbps).opus
I haven't used Windows in years. But from what I read here, fb2k outputs Opus files at 48000Hz.
So, how does foo_abx handles files with different sample rates?
Or is that irrelevant because the audio reaching the output device is mixed with a certain sample rate supported by the device?
can you nail down a freeformat@640?
If you mean "lame.exe --freeformat -b 640 *.wav", I cannot find how to decode it. Anyway, it must be useless.
I haven't used Windows in years. But from what I read here, fb2k outputs Opus files at 48000Hz.
So, how does foo_abx handles files with different sample rates?
Or is that irrelevant because the audio reaching the output device is mixed with a certain sample rate supported by the device?
If you decode it into wav, it will be 44kHz, so I think it is ok.
Musepack q9 (382 kbps)
(http://s9.postimage.org/b43bn6ri3/file.jpg) (http://postimage.org/image/b43bn6ri3/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/23 20:17:12
File A: D:\@Portable@\Audio\@Encoders@\test.wav
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\test (Musepack q9.0).mpc
20:17:12 : Test started.
20:19:18 : 01/01 50.0%
20:20:50 : 02/02 25.0%
20:21:50 : 03/03 12.5%
20:23:14 : 04/04 6.3%
20:27:34 : 05/05 3.1%
20:28:29 : 06/06 1.6%
20:29:12 : 07/07 0.8%
20:35:04 : 08/08 0.4%
20:35:52 : 08/09 2.0%
20:37:59 : 09/10 1.1%
20:41:08 : 10/11 0.6%
20:41:12 : Test finished.
----------
Total: 10/11 (0.6%)
No, Opus will not decode at 44100 Hz by default. 2012 is correct; please actually read what people say, rather than just waving it away.
I presume foo_abx intelligently resamples, probably both files to a common rate (the higher of the two?); I would certainly welcome confirmation.
Anyway: Thanks for all the tests! They make TOS #8 happy.
No, Opus will not decode at 44100 Hz by default. 2012 is correct; please actually read what people say, rather than just waving it away.
Opusdec will decode at 44.1kHz, if the opus file was sourced from a 44.1kHz WAV. Maybe that's what softrunner meant.
can you nail down a freeformat@640?
If you mean "lame.exe --freeformat -b 640 *.wav", I cannot find how to decode it.
fb2k cannot play it. I am not on my home computer right now, but ... tried lame.exe --decode and then comparing the WAV's? Though, according to http://mp3decoders.mp3-tech.org/freeformat.html (http://mp3decoders.mp3-tech.org/freeformat.html) , older Lame couldn't decode the highest-bitrate freeformats, but MAD can. http://www.underbit.com/products/mad/ (http://www.underbit.com/products/mad/)
Anyway, it must be useless.
Absolutely, except possibly for this particular purpose.
Opusdec will decode at 44.1kHz, if the opus file was sourced from a 44.1kHz WAV. Maybe that's what softrunner meant.
Exactly. Surprisingly, foobar2000 decodes at 48kHz.
... tried lame.exe --decode and then comparing the WAV's?
It looks like it works:
d:\@Portable@\Audio\@Encoders@> lame.exe --decode test_freeformat_640.mp3
Input file is freeformat.
hip: bitstream problem, resyncing skipping 2088 bytes...
input: test_freeformat_640.mp3 (44.1 kHz, 2 channels, MPEG-1 Layer III)
output: test_freeformat_640.wav (16 bit, Microsoft WAVE)
skipping initial 1105 samples (encoder+decoder delay)
skipping final 1015 samples (encoder padding-decoder delay)
Frame# 308/308 168 kbps MS
I'll try it later.
-----------------
To all. You may not believe me, but I've passed ABX test of Vorbis Q10 (619 kbps)
(http://s8.postimage.org/h0hqwgm8x/file.jpg) (http://postimage.org/image/h0hqwgm8x/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/24 00:58:35
File A: S:\320 GB HDD\Аудио-тесты\NEW\Converter\input\test.flac
File B: S:\320 GB HDD\Аудио-тесты\NEW\Converter\output\test (Vorbis q10,0).ogg
00:58:35 : Test started.
01:02:12 : 01/01 50.0%
01:04:00 : 01/02 75.0%
01:04:31 : 02/03 50.0%
01:05:51 : 03/04 31.3%
01:07:24 : 04/05 18.8%
01:08:09 : 05/06 10.9%
01:08:36 : 06/07 6.3%
01:08:59 : 07/08 3.5%
01:09:12 : 08/09 2.0%
01:10:02 : 09/10 1.1%
01:10:58 : 10/11 0.6%
01:11:39 : Test finished.
----------
Total: 10/11 (0.6%)
The difference in signals is that in the original sample the background flow of noise is not interrupted, but in Vorbis it is got interrupted somehow, other signals overlap it, so that I am loosing it (cannot hear it constantly).
can you nail down a freeformat@640?
I've met some difficulties with it. I know that the difference is there (original sounds sharper), but I can't get enought percent in abx test for posting it (as I remember, I had 9 of 10 but then made some mistakes). For me it is painfull to search this difference, so I'd better focus on testing Opus.
In Opus tests I got stuck at 200+ kbps (300+ real bitrate), but on portable player I recognize at least 224 kbps (converted in vorbis q10), so have to try it more.
I wonder how a hybrid format would perform on this. Do you mind trying wavpack -b256hx4 or even -b352hx4
LossyWAV, too.
lossyWAV (standard) yields a bitrate of 907kbit/s compared to FLAC 1013kbit/s [attached]....
I'm curious, what sound card do you have, softrunner? I remember back in the old days my Sound Blaster Audigy, that was also AC'97 compatible, mutilated the output enough to make lossy codecs much easier to ABX.
I wonder how a hybrid format would perform on this. Do you mind trying wavpack -b256hx4 or even -b352hx4
lossyWAV (standard) yields a bitrate of 907kbit/s compared to FLAC 1013kbit/s [attached]....
Can't pass it. For me they sound like lossless.
By the way I have ABXed Musepack q10 to be sure I have not lost my sensitivity due to listening the same sample that much.
Musepack Q10 (420 kbps)
(http://s12.postimage.org/3w0ng2dh5/file.jpg) (http://postimage.org/image/3w0ng2dh5/)
foo_abx 1.3.4 report
foobar2000 v1.1.14 beta 1
2012/07/27 04:02:32
File A: D:\@Portable@\Audio\LossyAudioTester\input\test.flac
File B: D:\@Portable@\Audio\LossyAudioTester\output\test (Musepack q10.0).mpc
04:02:32 : Test started.
04:16:17 : Trial reset.
04:17:35 : 01/01 50.0%
04:18:35 : 02/02 25.0%
04:20:18 : 02/03 50.0%
04:20:58 : 03/04 31.3%
04:21:56 : 04/05 18.8%
04:22:43 : 05/06 10.9%
04:23:38 : 06/07 6.3%
04:24:37 : 07/08 3.5%
04:26:31 : 08/09 2.0%
04:26:58 : 09/10 1.1%
04:29:16 : 10/11 0.6%
04:29:21 : Test finished.
----------
Total: 10/11 (0.6%)
(it sounds smoothed on the background in the right channel)
I'm curious, what sound card do you have, softrunner?
It is just standard AC97 (ALC850).
Opus 256 kbps (326 kbps) (git version from 10.11.2012)
(http://s14.postimage.org/rifklxyot/file.jpg) (http://postimage.org/image/rifklxyot/)
foo_abx 1.3.4 report
foobar2000 v1.1.16
2012/11/19 04:32:08
File A: D:\@Portable@\Audio\LossyAudioTester\output\FighterBeatLoop\FighterBeatLoop.flac
File B: D:\@Portable@\Audio\LossyAudioTester\output\FighterBeatLoop\FighterBeatLoop (Opus 256 kbps).opus
04:32:08 : Test started.
04:32:28 : 01/01 50.0%
04:32:53 : 02/02 25.0%
04:33:18 : 03/03 12.5%
04:35:05 : 03/04 31.3%
04:36:00 : 04/05 18.8%
04:37:36 : 05/06 10.9%
04:37:53 : 05/07 22.7%
04:38:14 : 06/08 14.5%
04:38:51 : 06/09 25.4%
04:39:56 : 07/10 17.2%
04:42:56 : 08/11 11.3%
04:45:31 : 09/12 7.3%
04:46:25 : 10/13 4.6%
04:47:15 : 11/14 2.9%
04:48:35 : Test finished.
----------
Total: 11/14 (2.9%)
Encoded sample sounds slightly poorer then the original one, with less amplitude of a saw-like sound.
Sorry to bump this, but this statement (kills all lossy codecs) looked too interesting.
Is anyone able to reproduce this now in 2018 with Opus, or any other lossy codecs?
I tested this sample with Opus (version: opus-tools 0.1.10 using libopus 1.2.1, parameters: "--padding 0 --bitrate 185") and Musepack (mppenc 1.30.0, default settings), resulting file size was 235151 for Opus and 234736 bytes for MPC, in neither case could I hear any differences, and it looks like Musepack was not even updated since 2009, and people were able to hear the difference even at MPC max quality settings, so does that mean I am just deaf and should give it up? Or is it okay no not hear this if I am 27 years old?
I also tried Opus --bitrate 128, it seems it's at the edge of my ability, it could be that the rhythmic scratching sound is made less sharp, but I am not completely sure I hear it or it might be guessing, in ABX testing training mode I got 7/7 (it took about 10 minutes of careful listening with too much loudness) and then I failed 8th try. I am not sure I am able to 100% correctly pass it to post a worthy log, and it's too painful.
The difference becomes really obvious with Opus --bitrate 64 (and the actual average bitrate is 82 then). So the transition for me must be somewhere in between.
Actually, I now found how to listen this. Now I can easily ABX it with QAAC (VBR, target bitrate 128) and Opus 128. In both cases (Opus and QAAC) the difference is that saw-like sound is made more "neutral" sounding after encoding.
Also tried MPC Q4, also ABX-able, but harder and the difference is different (instead of making scratching sound more neutral, it just adds some squeaking noises).
The trick is to know what to give attention to. Can be learned by adding the difference signal along with the original into a DAW (like Audacity) and trying with various gain for the difference track, muting and unmuting it. After a couple of minutes spent doing this, it becomes more clear how to find difference here, which may be unnoticeable at first.
A lot of people already passed this test I guess so I just did 8/8 and didn't save logs, but in case anyone needs it, I can try to pass 16/16 and post the logs.
relative rankings (how easy to detect blindly)
QAAC: easy
Opus: medium at first, became harder after getting tired a bit but taking a break fixes it
Musepack: hard (was not always sure that I pick the right thing, ended up 8/8 anyway)
Now, I'm just feeling stupid as I have to increase the bitrate for encoding music to portable devices or maybe even think about changing format again, and it'd suck because there isn't much space already...
Actually, tried again and Musepack is also affected just the same, also has the same kind of difference as QAAC and Opus.
Easy to ABX even at default preset (Q 5). So I can now get why some people said it "kills" "all" codecs.
So untimately it seems Opus is relatively the best here, hardest to ABX (but still not transparent at 128)
@magicgoose Have you tried encoding this sample with Opus @ vbr -140 ?
I've read before that most problematic samples became transparent or at least non-ABX-able at this bitrate.
I'll try Opus @ 140 later today maybe. Hopefully that will become transparent.
140 definitely harder than 128 but I think it's still not a guesswork yet.
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-19 14:09:29
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop opus 140.opus
SHA1: c229cced96be7836872632c1f3a5c09cd4ba5f0f
Gain adjustment: -6.16 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
14:09:29 : Test started.
14:10:14 : 01/01
14:10:20 : 02/02
14:10:41 : 03/03
14:10:52 : 04/04
14:11:11 : 05/05
14:11:33 : 06/06
14:11:51 : 07/07
14:12:01 : 08/08
14:12:01 : Test finished.
----------
Total: 8/8
p-value: 0.0039 (0.39%)
-- signature --
72720452143acc87b189f6ab7fb3f2dce3349f64
8/8 even for Opus @ 200. But it was very hard, and I wasn't always sure that I definitely hear it or it's in my head.
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-19 14:17:10
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop opus 200.opus
SHA1: 7b7090d4acbb266a77f2480696d901978d3972ae
Gain adjustment: -6.17 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
14:17:10 : Test started.
14:18:29 : 01/01
14:18:52 : 02/02
14:19:03 : 03/03
14:19:17 : 04/04
14:19:41 : 05/05
14:19:59 : 06/06
14:20:05 : 07/07
14:20:16 : 08/08
14:20:16 : Test finished.
----------
Total: 8/8
p-value: 0.0039 (0.39%)
-- signature --
fa735075f5989a252419922560a29be2a83e8223
Tried with MPC @ Q7, completely unsure if I can hear the difference, ABX failed. Guess I'll settle for it for now, overkill yes, but I don't know any samples that kill it, and it's still not as bad as using stuff like lossywav / wavpack.
(but I can't be sure that I won't hear it at another day when I'm less tired... if only I could have enough space on portable devices to not care about lossy, lol)
So this fighter loop is really tough one. :D
Which setting did you use on wavpack lossy?
I didn't test wavpack lossy.
But you said in previous post:
"but I don't know any samples that kill it, and it's still not as bad as using stuff like lossywav / wavpack"
What did you mean by that?
I thought that you have tested wavpack before. Probably missed something...
I wonder how a hybrid format would perform on this. Do you mind trying wavpack -b256hx4 or even -b352hx4
This would definitely be an interesting test. If someone is willing to do it. :)
> What did you mean by that?
sorry, I didn't mean it.
I meant that using lossywav or wavpack for all the music would probably be very inefficient due to the nature of these.
Wavpack doesn't even have "true" VBR mode (where the deciding factor is a quality level and not bitrate).
Lossywav + FLAC maybe is actually still worth trying, I'll maybe try it later. But it's still not very practical choice even compared to higher presets of MPC:
https://wiki.hydrogenaud.io/index.php?title=LossyWAV
The minimum level which is expected to at least "likely to be transparent" gives 407 kbit/s on the "10 Album Test Set", that's really a lot, even MPC/Q10 is expected to give "only" 350 on average
Wavpack doesn't even have "true" VBR mode (where the deciding factor is a quality level and not bitrate).
That's true however, wavpack lossy @ -b350hx4 should yield very high quality.
MPC Q7 ABX
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 13:58:13
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop mpc q7.mpc
SHA1: d648e0a463cb6817af7c7dafedf13f605c6a4e30
Gain adjustment: -6.17 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
13:58:13 : Test started.
13:58:59 : 01/01
13:59:22 : 02/02
13:59:35 : 03/03
13:59:46 : 04/04
13:59:56 : 05/05
14:00:13 : 06/06
14:00:35 : 07/07
14:02:31 : 08/08
14:03:02 : 09/09
14:03:20 : 10/10
14:03:36 : 11/11
14:05:44 : 12/12
14:07:09 : 13/13
14:08:27 : 14/14
14:08:34 : 14/15
14:09:26 : 15/16
14:09:26 : Test finished.
----------
Total: 15/16
p-value: 0.0003 (0.03%)
-- signature --
ac7d37d206823e19ce9b5e287afb6f34a27d9836
Quickly got tired and it became even harder at the last iterations, I was surprised to get 15/16 because I thought it'd be lower
So, probably MPC Q7 also isn't enough.
Try MPC with max mid side mode; --ms 15 (default for Q10):
--quality 7 --ms 15
--quality 7.5 --ms 15
Apple AAC (Q 100) - Very easy (after figuring out which part of the sound goes missing after encoding, it's easy to pick it up)
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 14:17:54
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop aac apple q100.m4a
SHA1: af6b98b25e262f1661f971ba2a6e53a0f018bcee
Gain adjustment: -6.15 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
14:17:54 : Test started.
14:28:49 : 01/01
14:29:15 : 02/02
14:29:30 : 03/03
14:29:49 : 04/04
14:30:02 : 05/05
14:30:14 : 06/06
14:30:26 : 07/07
14:30:40 : 08/08
14:30:48 : 09/09
14:30:56 : 10/10
14:31:04 : 11/11
14:31:13 : 12/12
14:31:23 : 13/13
14:31:36 : 14/14
14:31:49 : 15/15
14:32:01 : 16/16
14:32:01 : Test finished.
----------
Total: 16/16
p-value: 0 (0%)
-- signature --
846bd78466866f867eef7a7fa90b812eeced24d1
Also found that (at least for Apple AAC) the difference is easiest to hear not at the very big loudness, but something closer to medium.
If made louder, it begins to overwhelm ears somehow and I get lost.
Actually MPC Q7 is just as easy, after focusing on the same sound, training indeed does a lot
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 14:41:53
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop mpc q7.mpc
SHA1: d648e0a463cb6817af7c7dafedf13f605c6a4e30
Gain adjustment: -6.17 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
14:41:53 : Test started.
14:43:07 : 01/01
14:43:19 : 02/02
14:43:30 : 03/03
14:43:40 : 04/04
14:43:56 : 05/05
14:44:08 : 06/06
14:44:19 : 07/07
14:44:33 : 08/08
14:44:57 : 09/09
14:45:11 : 10/10
14:45:42 : 11/11
14:45:55 : 12/12
14:46:01 : 13/13
14:46:15 : 14/14
14:46:26 : 15/15
14:46:38 : 16/16
14:46:38 : Test finished.
----------
Total: 16/16
p-value: 0 (0%)
-- signature --
c1ff9bceefcc8582bd7f8a3ead06bb9c2d2895f4
(file same as before when I did 15/16 and with a lot of tryharding)
Try MPC Q7 or Q7.5 with --ms 15 switch
That's what I'm trying right now. Trying to recall the shape of that sound, I always forget it and need to find it again.
But I already see that the --ms switch alone changes the file size by only ~200 bytes. (It doesn't decode identically to the plain Q7 file).
I'm trying Q7.5 --ms 15, if I can't, will try the same for Q7
after recalling what to focus on, Q7.5 with that extra switch is just about as easy as Q7
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 15:00:33
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop mpc q7.5 ms 15.mpc
SHA1: ce2a7cb1d2315e60978696769c79a57db98c1fac
Gain adjustment: -6.18 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
15:00:33 : Test started.
15:01:21 : 01/01
15:01:33 : 02/02
15:01:43 : 03/03
15:01:53 : 04/04
15:02:05 : 05/05
15:02:16 : 06/06
15:02:27 : 07/07
15:02:41 : 08/08
15:02:55 : 09/09
15:03:05 : 10/10
15:03:17 : 11/11
15:03:32 : 12/12
15:03:49 : 13/13
15:04:09 : 14/14
15:04:26 : 15/15
15:04:36 : 16/16
15:04:36 : Test finished.
----------
Total: 16/16
p-value: 0 (0%)
-- signature --
6c5f6e32dce5417d82199bba50d2fa57905c19c3
Are you willing to try wavpack lossy using -b350hx4 ?
Are you willing to try wavpack lossy using -b350hx4 ?
Maybe later.
--
Decided to try mp2 at max bitrate (twolame -b 384), it's also easy to ABX.
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 16:56:57
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop.mp2
SHA1: ae9d51c69bfb79c0ff862e924336486ff236f0da
Gain adjustment: -6.17 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
16:56:57 : Test started.
16:59:53 : 01/01
17:00:04 : 02/02
17:00:20 : 03/03
17:00:31 : 04/04
17:00:42 : 05/05
17:00:51 : 06/06
17:01:04 : 07/07
17:01:31 : 08/08
17:01:49 : 09/09
17:01:58 : 10/10
17:02:08 : 11/11
17:02:24 : 12/12
17:02:35 : 13/13
17:02:50 : 14/14
17:03:14 : 15/15
17:03:32 : 16/16
17:03:32 : Test finished.
----------
Total: 16/16
p-value: 0 (0%)
-- signature --
3a5fe1fa18a07a9f4678a46dca66ed1dd40bbc12
Are you willing to try wavpack lossy using -b350hx4 ?
Can't hear difference, but it ends up using 476kbps for this sample with these settings.
MPC Q10 still ABX-able
foo_abx 2.0.6c report
foobar2000 v1.4.6
2019-08-20 17:12:37
File A: Fighter_Beat_Loop.flac
SHA1: 796f85a948de5135e0393930694ee0fb4c487b00
Gain adjustment: -6.18 dB
File B: Fighter_Beat_Loop mpc q10.mpc
SHA1: 5e32c8d414d9dc472ff9bbb3ac2185cb6c17c774
Gain adjustment: -6.18 dB
Output:
WASAPI (shared) : Out: default
Crossfading: NO
17:12:37 : Test started.
17:13:42 : 01/01
17:13:53 : 02/02
17:14:33 : 03/03
17:14:42 : 04/04
17:14:52 : 05/05
17:15:17 : 06/06
17:15:24 : 07/07
17:15:46 : 08/08
17:16:04 : 09/09
17:16:24 : 10/10
17:16:38 : 11/11
17:17:05 : 12/12
17:17:33 : 13/13
17:18:04 : 14/14
17:18:28 : 15/15
17:18:44 : 16/16
17:18:44 : Test finished.
----------
Total: 16/16
p-value: 0 (0%)
-- signature --
5e1d03d432660fa52380c071b99b44d62eaca318
Wondering if there are any possible "expert" settings to push past this level to really force it to keep that sound
You can try tweaking --nmt --tmn. nmt inflates bitrate more compared with tmn. You can see default values for each profile with --verbose
e.g:
--nmt 20 --tmn 40
mppenc --xtreme or --insane --nmt 20 --tmn 40
mppenc --quality 10 --nmt 20 --tmn 40
Are you willing to try wavpack lossy using -b350hx4 ?
Can't hear difference, but it ends up using 476kbps for this sample with these settings.
Thanks a lot. LossyWav/Flac standard ends up at 907k so this wavpack setting is much better.
Good news: these parameters for musepack are actually letting it scale all the way up to very small differences that have to be inaudible. (and it can take up to 931 bitrate if I put a ridiculously high value of 60dB into both)
Now I need to find the optimal setting which removes the artifact yet doesn't inflate the bitrate too much.
I am guessing that I have to focus on `--nmt`, but not 100% sure yet.
Also interesting if Opus has any similar fine tuning options?
Not sure just yet, might have to try again later, but I can't hear difference with MPC --quality 5 --nmt 18.
It gives bitrate 321kbps on this sample, not too bad yet. But maybe it is still overkill.
Perhaps it could have been even better if NMT could be tuned per band, it doesn't make sense to raise it for anything above 16kHz for example,
On "normal" music MPC "--quality 5 --nmt 18" ends up using from 190 to 296 kbps (tested on several albums picked from my collection trying to be as diverse as possible in terms of sound characteristics), so if the other choice is to use lossyflac or wavpack hybrid, this one is clearly more economic.
I probably should try to find the point when Opus becomes transparent here for a fair comparison, and also try to nail down a cheaper setting for MPC too.
Very interesting testing. It would be also interesting how this MPC setting deal with other problematic samples like badvilbel, eig, castanets etc. This could be THE setting to use in general.
I also tested this setting to see bitrate on various cd albums and can confirm is 240-250k on average. Very similar in bitrate like MPC q7-8 but with nmt 18 should be better.
I usually was not able to ABX MPC on any other samples that I know of, it would be interesting to let some people with better ears to test this setting on other samples.
Also: tested with NMT 17 and lower, it was ABX-able, it really only gets hard or impossible for me on NMT 18. Maybe it's somehow related to the fact that in standard preset, 18 is also the value for the other main parameter (TMN) and it should let encoder treat tonal and "noisy" content the same.
Not sure it's worth the trouble to search for a fractional value of NMT.
Would be interesting if there was some other parameter to tune the behavior of deciding what is tone and what is noise. Maybe that could make it possible to "waste" less bits on this sample to make it transparent.
I would try the WV hybrid route starting around 256k. One collection. On the PC your music becomes lossless through the wvc file and theres zero degradation possible. On portables you could copy the album folder and playback should also be lossless - assuming the player supports .wvc files. If space becomes scarce you can delete or not copy wvc in the first place - the same codec / collection becomes lossy saving 3 ~ 4 times space over lossless. I use -b3hx4cl and previously -b3x4cl without issues so far on. On PC its lossless, on my Samsung J2 pro (poweramp player) I copy only the WV files. Bitrate is around 270k.
Other reason, The hybrids can be immune to certain type of problem that affects traditional codecs (though things can also go opposite direction)
I usually was not able to ABX MPC on any other samples that I know of, it would be interesting to let some people with better ears to test this setting on other samples.
Also: tested with NMT 17 and lower, it was ABX-able, it really only gets hard or impossible for me on NMT 18. Maybe it's somehow related to the fact that in standard preset, 18 is also the value for the other main parameter (TMN) and it should let encoder treat tonal and "noisy" content the same.
Not sure it's worth the trouble to search for a fractional value of NMT.
Would be interesting if there was some other parameter to tune the behavior of deciding what is tone and what is noise. Maybe that could make it possible to "waste" less bits on this sample to make it transparent.
I'd use --extreme or --quality 6 with tweaks as its more suited for that in theory anyway. Maybe nmt 17 or lower would work.
I'd use --extreme or --quality 6 with tweaks as its more suited for that in theory anyway. Maybe nmt 17 or lower would work.
--quality 10 uses NMT 14 by default and that is ABX-able, so there isn't a whole lot of space between NMT 14 and 18 anyway.
I may guess that this will just end up using more bitrate, and it's not clear how that would improve anything except stuff that's already transparent.
I think that using setting --quality 5 + nmt 18 should be very good in general with bitrate in 250k range (and as you said very hard or impossible to ABX using nmt18) while default --quality 10 is in 350k range and ABX-able.
Hi,
Is someone willing to ABX badvilbel /eig samples with this setting (quality 5 -nmt 18)?
I'm not trained to spot artifacts and I don't have good equipment but
if this setting provides transparent, or very hard to ABX results, on critical samples,
I think this could be very interesting discovery for lossy formats in general while retaining "economical" bitrate of 240-270k on average using this setting.
Just thinking... :)
those are mp3 samples. mpc is less affected by standard codec issue like preecho. It has some edge cases here and there.
But I've read before that badvilbel is tough on varios codecs/formats. So using mpc Q5 -nmt 18 should yield very high if not transparent quality as magicgoose and you discovered on this fighter beat sample.