Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Results of trascoding test to 128k (Read 18394 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Results of trascoding test to 128k

I finished a short but interesting test today. My aim was to test the effects of high bitrate lossy transcoding using conventional codecs like lame mp3 and also to test the hybrid codecs such as wavpack.

Task is simple: transcode a large lossy file into a 128k mp3 so it will be suitable for portable players. I chose lame 3.96.1 APS, wavpack lossy around 400k and freeformat mp3 at 384k. I chose freeformat to test its reliability considering the popularity of allofmp3.com. These files were transcoded into 128k mp3 cbr using lame 3.96.1 - reference is 128k mp3. 128k is still popular with the masses and makes it easier to perform tests like these. The sample I chose was the wicked 'amnesia' .

This test was done casualy in a normal enviroment.

Now some results:

1) Mp3 128k transc from Freeformat 384k vs. original 128k mp3

ile A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia cbr128 ff384.mp3

15:56:04 : Test started.
15:57:18 : 01/01  50.0%
15:58:10 : 01/02  75.0%
15:58:58 : 01/03  87.5%
15:59:58 : 01/04  93.8%
16:00:55 : 02/05  81.3%
16:00:59 : Trial reset.
16:03:01 : 01/01  50.0%
16:04:07 : 02/02  25.0%
16:04:19 : 03/03  12.5%
16:04:32 : 03/04  31.3%
16:04:58 : 04/05  18.8%
16:05:04 : 04/06  34.4%
16:05:13 : 04/07  50.0%
16:05:22 : 05/08  36.3%
16:05:27 : 06/09  25.4%
16:05:32 : 06/10  37.7%
16:06:47 : Trial reset.
16:06:54 : 01/01  50.0%
16:07:04 : 02/02  25.0%
16:07:14 : 03/03  12.5%
16:07:20 : 04/04  6.3%
16:07:29 : 05/05  3.1%
16:07:46 : 05/06  10.9%
16:07:51 : 05/07  22.7%
16:08:00 : 05/08  36.3%
16:08:06 : 05/09  50.0%
16:08:11 : 05/10  62.3%
16:09:32 : 06/11  50.0%
16:09:38 : Trial reset.
16:09:43 : 01/01  50.0%
16:09:47 : 02/02  25.0%
16:09:52 : 03/03  12.5%
16:10:01 : 04/04  6.3%
16:10:11 : 05/05  3.1%
16:10:31 : 06/06  1.6%
16:10:39 : 07/07  0.8%
16:10:43 : 08/08  0.4%
16:10:48 : 09/09  0.2%
16:10:52 : 10/10  0.1%
16:10:57 : 11/11  0.0%
16:11:01 : 12/12  0.0%
16:11:34 : 13/13  0.0%
16:11:39 : 14/14  0.0%
16:11:43 : 15/15  0.0%
16:11:47 : 16/16  0.0%
16:11:51 : Test finished.

----------
Total: 30/42 (0.4%)


1.7-3.8 secs is were I found bad noises and dropouts. (16/16)


2) 128k mp3 transc from Lame APS 3.96.1 vs. original 128k mp3

File A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia 128  transc aps .mp3

17:01:15 : Test started.
17:02:08 : 01/01  50.0%
17:02:16 : 02/02  25.0%
17:02:22 : 03/03  12.5%
17:02:29 : 04/04  6.3%
17:02:49 : 04/05  18.8%
17:02:56 : 05/06  10.9%
17:03:05 : 06/07  6.3%
17:03:52 : 07/08  3.5%
17:03:59 : 08/09  2.0%
17:04:05 : 09/10  1.1%
17:04:16 : 10/11  0.6%
17:04:41 : Test finished.

----------
Total: 10/11 (0.6%)


1.7-3.8 secs: Better  than freeformat ! ?? but new ringing / smearing. Overall much less noisy than freeformat.

Another trial with same files (1st 3 secs):

File A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia 128  transc aps .mp3

16:24:09 : Test started.
16:26:47 : 01/01  50.0%
16:26:55 : 01/02  75.0%
16:27:10 : Trial reset.
16:27:30 : 01/01  50.0%
16:27:38 : 02/02  25.0%
16:27:51 : 03/03  12.5%
16:28:03 : 04/04  6.3%
16:28:11 : 04/05  18.8%
16:28:43 : 05/06  10.9%
16:29:11 : 06/07  6.3%
16:29:57 : 07/08  3.5%
16:30:05 : 08/09  2.0%
16:30:51 : 09/10  1.1%
16:31:13 : 10/11  0.6%
16:31:47 : 10/12  1.9%
16:31:57 : 11/13  1.1%
16:32:05 : 12/14  0.6%
16:32:18 : Test finished.

----------
Total: 13/16 (1.1%)

Fist 2 secs HF ringing


3) mp3 128k transc from wavpack 400k vs. original 128k mp3:

File A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia128 transc wv400.mp3

16:40:03 : Test started.
16:40:33 : 00/01  100.0%
16:40:54 : 01/02  75.0%
16:40:58 : 01/03  87.5%
16:41:05 : 02/04  68.8%
16:41:09 : 03/05  50.0%
16:41:14 : 03/06  65.6%
16:41:20 : 04/07  50.0%
16:41:24 : 05/08  36.3%
16:41:29 : 06/09  25.4%
16:41:33 : 06/10  37.7%
16:41:44 : 06/11  50.0%
16:41:51 : 06/12  61.3%
16:41:56 : Test finished.

----------
Total: 6/12 (61.3%)


The 1.7-3.8 secs are transparent to me.

Another trial  - 1st few secs:

File A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia128 transc wv400.mp3

16:34:17 : Test started.
16:34:24 : 01/01  50.0%
16:34:28 : 02/02  25.0%
16:34:35 : 02/03  50.0%
16:34:41 : 02/04  68.8%
16:34:45 : 03/05  50.0%
16:34:52 : 03/06  65.6%
16:35:12 : 04/07  50.0%
16:35:21 : 04/08  63.7%
16:35:32 : 04/09  74.6%
16:36:02 : 05/10  62.3%
16:36:19 : 05/11  72.6%
16:36:29 : 06/12  61.3%
16:36:34 : 07/13  50.0%
16:36:38 : 07/14  60.5%
16:36:45 : 08/15  50.0%
16:36:49 : 09/16  40.2%
16:36:55 : 10/17  31.5%
16:37:02 : 11/18  24.0%
16:37:14 : 11/19  32.4%
16:37:25 : 11/20  41.2%
16:37:27 : Test finished.

----------
Total: 11/20 (41.2%)

1st secs cannot abx hf boost / ringing.


Conclusion: Amnesia is a tough sample. transcoding from APS introduces hf ringing distortions. Freeformat was a big disappointment as there was less ringing but much worse noise than APS !.

Transcoding from high bitrate mp3 to 128k is questionable and with this sample its bad news.

Wavpack showed no audiable loss to me here. I am going to do more tests soon, but for now I can say that hybrid lossy around 400k or maybe lower can be used for general archiving and transcoding when space is short or you just don't like 870k lossless files. I am not advocating it as a replacement for true lossless though but with the correction files you might get the best of both.

Results of trascoding test to 128k

Reply #1
Quote
Freeformat was a big disappointment as there was less ringing but much worse noise than APS !.
[a href="index.php?act=findpost&pid=230764"][{POST_SNAPBACK}][/a]


Was this evident in the original freeformat encoding, aswell as the transcoded sample?

Thanks for sharing your findings with us.
daefeatures.co.uk

Results of trascoding test to 128k

Reply #2
I couldn't abx the freeformat vs FLAC 'amnesia' in the 1.7-3.8 secs.

Results of trascoding test to 128k

Reply #3
Interesting test. Thanks for your effort.
"To understand me, you'll have to swallow a world." Or maybe your words.

Results of trascoding test to 128k

Reply #4
I did some more similar tests using same scenario with Vorbis Aotuvb2.

a) Vorbis Q6 transcoded to Q0 for portable usage. Transcoded file vs reference file:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 22:51:43

File A: file://H:\temp\abx tests\amnesia ogg 0 ref.ogg
File B: file://H:\temp\abx tests\amnesia ogg 0 trans from ogg 6.ogg

22:51:44 : Test started.
22:52:21 : 01/01  50.0%
22:52:26 : 01/02  75.0%
22:52:30 : 02/03  50.0%
22:52:36 : 02/04  68.8%
22:52:41 : 03/05  50.0%
22:52:47 : 03/06  65.6%
22:52:56 : 04/07  50.0%
22:53:08 : 05/08  36.3%
22:53:21 : 06/09  25.4%
22:53:29 : 07/10  17.2%
22:53:36 : 08/11  11.3%
22:53:53 : 09/12  7.3%
22:54:26 : 10/13  4.6%
22:54:59 : 11/14  2.9%
22:55:16 : 12/15  1.8%
22:56:07 : 13/16  1.1%
22:56:28 : 13/17  2.5%
22:56:41 : 14/18  1.5%
22:56:49 : 15/19  1.0%
22:56:54 : 16/20  0.6%
22:57:08 : Test finished.

----------
Total: 16/20 (0.6%)

noise, hf boost


b) Lame 3.96.1 APS transcoded to Vorbis Q0. transcoded file vs reference.

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 22:28:47

File A: file://H:\temp\abx tests\amnesia ogg 0 ref.ogg
File B: file://H:\temp\abx tests\amnesia ogg 0 trans aps.ogg

22:28:49 : Test started.
22:29:31 : 01/01  50.0%
22:29:37 : 02/02  25.0%
22:29:43 : 03/03  12.5%
22:29:48 : 04/04  6.3%
22:30:00 : 05/05  3.1%
22:30:05 : 06/06  1.6%
22:30:09 : 07/07  0.8%
22:31:03 : 07/08  3.5%
22:31:10 : 08/09  2.0%
22:31:18 : 09/10  1.1%
22:31:33 : 10/11  0.6%
22:31:36 : Test finished.

----------
Total: 10/11 (0.6%)

overall HF noise


c) Vorbis Q6 transcoded to Lame 3.96.1 128k cbr. transcoded mp3 vs reference 128k mp3 :

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 18:21:28

File A: file://H:\temp\abx tests\amnesia128cbr.mp3
File B: file://H:\temp\abx tests\amnesia 128 lame trans ogg q6.mp3

18:21:29 : Test started.
18:21:38 : 01/01  50.0%
18:21:56 : 02/02  25.0%
18:22:31 : 03/03  12.5%
18:25:26 : 04/04  6.3%
18:25:34 : 05/05  3.1%
18:26:01 : 05/06  10.9%
18:26:15 : 06/07  6.3%
18:26:22 : 07/08  3.5%
18:26:39 : 08/09  2.0%
18:26:51 : 09/10  1.1%
18:27:03 : 10/11  0.6%
18:27:31 : 11/12  0.3%
18:27:39 : 12/13  0.2%
18:27:53 : 13/14  0.1%
18:28:35 : 14/15  0.0%
18:28:48 : Test finished.

----------
Total: 14/15 (0.0%)

hf distortion


d) Wavpack hybrid lossy using high quality 320k target bitrate (-hb320) transcoded to Vorbis Q0. Reference vorbis Q0 vs transcoded file.

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 23:09:50

File A: file://H:\temp\abx tests\amnesia ogg 0 ref.ogg
File B: file://H:\temp\abx tests\amnesia ogg 0 transc from wv320.ogg

23:09:52 : Test started.
23:10:20 : 00/01  100.0%
23:10:31 : 00/02  100.0%
23:10:49 : 01/03  87.5%
23:11:00 : 01/04  93.8%
23:11:06 : 02/05  81.3%
23:11:14 : 03/06  65.6%
23:11:19 : 04/07  50.0%
23:11:25 : 04/08  63.7%
23:11:31 : 04/09  74.6%
23:11:36 : 05/10  62.3%
23:11:42 : 05/11  72.6%
23:11:50 : 05/12  80.6%
23:12:01 : 05/13  86.7%
23:12:30 : 05/14  91.0%
23:12:44 : 06/15  84.9%
23:13:14 : 06/16  89.5%
23:13:34 : Test finished.

----------
Total: 6/16 (89.5%)

Another go:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 22:41:44

File A: file://H:\temp\abx tests\amnesia ogg 0 ref.ogg
File B: file://H:\temp\abx tests\amnesia ogg 0 transc from wv320.ogg

22:41:45 : Test started.
22:42:00 : 00/01  100.0%
22:42:16 : 00/02  100.0%
22:42:30 : 00/03  100.0%
22:43:04 : 00/04  100.0%
22:43:16 : 01/05  96.9%
22:43:21 : 02/06  89.1%
22:43:27 : 02/07  93.8%
22:43:37 : 02/08  96.5%
22:43:41 : 02/09  98.0%
22:43:44 : Test finished.

----------
Total: 2/9 (98.0%)

No luck with artifacts. I will try to rapidly abx background hiss:

foo_abx v1.2 report
foobar2000 v0.8.3
2004/08/17 23:02:34

File A: file://H:\temp\abx tests\amnesia ogg 0 ref.ogg
File B: file://H:\temp\abx tests\amnesia ogg 0 transc from wv320.ogg

23:02:35 : Test started.
23:02:48 : 01/01  50.0%
23:02:53 : 02/02  25.0%
23:03:01 : 03/03  12.5%
23:03:07 : 03/04  31.3%
23:03:12 : 04/05  18.8%
23:03:16 : 04/06  34.4%
23:03:24 : 05/07  22.7%
23:03:29 : 05/08  36.3%
23:03:36 : 06/09  25.4%
23:03:51 : 07/10  17.2%
23:04:01 : 08/11  11.3%
23:04:08 : 08/12  19.4%
23:04:16 : 09/13  13.3%
23:04:20 : 10/14  9.0%
23:04:28 : 11/15  5.9%
23:04:42 : 11/16  10.5%
23:05:35 : Test finished.

----------
Total: 11/16 (10.5%)

Quite difficult. Feels more like imagination.

Results of trascoding test to 128k

Reply #5
Quote
I chose freeformat to test its reliability considering the popularity of allofmp3.com.[a href="index.php?act=findpost&pid=230764"][{POST_SNAPBACK}][/a]

Good work here. By the way, I've seen freeformat 384 kbps mentioned before in regards to allofmp3, including in a popular review out there. I signed up just for the heck of it to see what all the fuss was about (after it was mentioned on The Screensavers), and noticed that when you order sutff using their Online Encoding feature, it will tell you the source format of the file you're about to encode. 99% of the time it says -alt-preset standart (the t is a typo obviously). If true that would mean very little of their music is actually in 384 kbps freeformat. Bad news for transcoding. Makes me wonder how much of their lossless stuff can be trusted as well. The problem is you just don't know where the source came from. Plus you have no control over the ripping process (are all their songs listend to for errors?). In the end I decided I didn't really care for the service, despite it being so cheap (and yes, I know allofmp3 is a grey area legally speaking). I'd much rather be able to buy songs from a reputable store in a lossless format. Anyways, thanks for the tests. Interesting stuff.

Results of trascoding test to 128k

Reply #6
Thanks for the test.

While it was carried out quite differently from some tests I posted a year ago or so, the findings were similar. In my case, I was looking for a source lossy format for transcoding into ATRAC for Minidisc. Wavpack was much better than any of the other lossy formats at the time for this. (I have since tested Optifrog Dualstream which also works trasncodes well.)

It's always reassuring to see others get similar results.

Den.

Results of trascoding test to 128k

Reply #7
could you try (just for fun) a transcode test between a subband codec and a transform.

i have a little theory (with no scientific basis) that transcodes from subband to transform (and possibly vice-versa) will stand up better than transform to transform or sub to sub.

i should do it myself, but i'm a tad lazy at the moment

a good place to start would be musepack standard to lame --aps.

Results of trascoding test to 128k

Reply #8
I encoded the track 'take my breath away' by Berlin to LAME 3.96.1 V7 from FLAC and from MPC --insane. I focused on a 3 second drum intro. They seemed very close but hit a,b,y a few times and there is something wrong with the cymbals. They sounded smeared and lost sharpness.

mpc --insane : 7/7, 2/2

wavpack -b320x : 12/16, 5/8  - too difficult.

Re: Results of trascoding test to 128k

Reply #9
I encoded the track 'take my breath away' by Berlin to LAME 3.96.1 V7 from FLAC and from MPC --insane. I focused on a 3 second drum intro. They seemed very close but hit a,b,y a few times and there is something wrong with the cymbals. They sounded smeared and lost sharpness.

mpc --insane : 7/7, 2/2

wavpack -b320x : 12/16, 5/8  - too difficult.

Very interesting that Wavpack at even -b320x is enough for this song.

Are you willing to test MPC Quality 5 --nmt 18 transcoded to Mp3 128 ABR (or V5)?
I think it should be better than mpc insane.

 

Re: Results of trascoding test to 128k

Reply #10
Quote
(and yes, I know allofmp3 is a grey area legally speaking).
Since the artists and other copyright holders are obviously not getting paid, you are essentially paying someone to steal music for you.