How Good Is Transcoding?

2002-09-26 15:50:40

I was wondering about transcoding with AAC..  I store songs I can't get at in either -normal or -extreme, but if I ever got a hardware player I'd want to make smaller copies to slap on my player.  I was wondering, what steps down can I take without losing quality?

For example, is going from:  wav -> extreme -> internet  a big enough step down for there not to be any problem..

I already know I'm not going to be able to ABX myself, the only way I've found to judge the quality of a file is by looking at the spectrum in CoolEd.  I don't know why, but I can't even ABX an original wave from one encoded at -internet (112kbs)  .. [span style='font-size:8pt;line-height:100%'](or  considering it means internet is transparent encoding!)[/span]



So, here's what I've done.. 

[span style='font-size:8pt;line-height:100%']Taken three 20-30 second samples, metal, electronic, and classical.  They probably aren't the best samples to take, but oh well

All encoding is with Psytel 2.15, and decoding with the Cool Edit plugin (dated 22/8)

I encoded them in -extreme, then decoded and encoded in -streaming
Then to compare, I encoded them in -streaming.

Then I did the same, encoding in -normal, then decoded and encoded in -internet. Then compared by encoding in -internet.

The only changes I've noticed are in the final bitrates.  I can't tell any difference between the files..  only that the transcoded ones end up a slightly higher bitrate, which I'm guessing is the encoder doing its job, making them the specified level of quality.

The bitrate changes were (internet->normal to internet transcode)
114->125, 88->89, 112->118
and (streaming->extreme to streaming transcode)
140->145, 104->103??, 137->139

[/span]

So the test is, can you tell the difference between the regular and transcoded samples?  The transcodings have both gone down 2 steps, which I am hoping is enough of a drop to make little or no difference.

If anyone who's interested and has better hearing than me is interested, please grab these files [span style='font-size:8pt;line-height:100%'](FLAC's, which I saw should be used)[/span] and post your ABX results back. 

extreme -> streaming test
Korn - Blind (2.9mb each)
Original
Transcode

Fatboy Slim - Acid 8000 (2.1mb each)
Original
Transcode

Erik Satie - Gymnopedies (2.5mb each)
Original
Transcode

normal -> internet test
Korn - Blind (2.8mb each)
Original
Transcode

Fatboy Slim - Acid 8000 (2.0mb each)
Original
Transcode

Erik Satie - Gymnopedies (2.3mb each)
Original
Transcode

Anyone who tries some of these, many thanks to you!  Hopefully this'll find something out.

Mac

How Good Is Transcoding?

Reply #1 – 2002-09-27 00:16:15

Yeah, this is pretty interesting test..
I'll try it shortly. But my guess is that it's pretty hard to tell the difference at least between extreme->streaming and original wav->streaming. Actually I just tested the first pair quickly in winamp and at least couldn't immediately tell which is the transcoded and which is the "virgin" encode. Gonna try seriously using ABA (less trials  ) tomorrow.

How Good Is Transcoding?

Reply #2 – 2002-09-27 10:09:10

Thanks  That's what I'm hoping for, if you can't tell the difference between them, the only sacrifice is a ~5kbs bigger file

How Good Is Transcoding?

Reply #3 – 2002-09-27 10:54:36

This is interesting
I gotta solder up my headphones.. fast

How Good Is Transcoding?

Reply #4 – 2002-10-02 13:00:46

If anyone is going to test.. could they do so in the next week or so please?  I've had such a weak response I'm gonna take the files down pretty shortly, they're just taking up space on a friends webspace!

How Good Is Transcoding?

Reply #5 – 2002-10-02 13:10:45

Yeah, I'll do this test I promise.
I donwloaded the samples, and will possibly try this tonight. Hopefully some other people try this too...

How Good Is Transcoding?

Reply #6 – 2002-10-03 22:55:02

extreme -> streaming test
Korn - Blind (2.9mb each)

-------------------------------------
WinABX v0.22 (ABA) test report
10/04/2002 00:18:44

A file: F:\AAC Transcode test\KoRn-s137.wav
B file: F:\AAC Transcode test\KoRn-x234-s139.wav

00:24:14    1/1  p=33.3%
00:24:51    2/2  p=11.1%
00:25:33    3/3  p= 3.7%
00:25:45    4/4  p= 1.2%
00:25:59    5/5  p= 0.4%
00:30:57  reset

Double encoded file had attenueated high freq squeek near 4.7s while it's more audible in the virgin encode.



00:31:59    1/1  p=33.3%
00:32:41    2/2  p=11.1%
00:33:18    3/3  p= 3.7%
00:33:44    4/4  p= 1.2%
00:35:19    5/5  p= 0.4%

9.6s - 12.8s cymbals in the double encode are fluttering/more watery sounding.



00:40:12    0/1  p=100.0%
00:40:33    1/2  p=55.5%
00:40:59    2/3  p=26.0%
00:42:52    3/4  p=11.1%
00:43:15    4/5  p= 4.5%
00:43:52    5/6  p= 1.8%
00:44:35    6/7  p= 0.7%

The big drum hit at about 20.8 is a bit more rough with the double encode.
----------------------------------------
About the quality of double encode vs virgin: ok, it was pretty easy to find places where it's easy to ABA once you are a bit familiar with the track. If you are listening the first time and the track is not familiar to you, then it's quite hard to notice much difference, but once I tried seriously, it wasn't particularly hard to hear differencies.

So this was the first pair. More to come..