Double blind test 128Kbps vs lossless? I'll be amazed if you can tell much difference - Page 7

The recordings, to me, are quite similar but I prefer recording number two because the attacks of the strings in number two seem more open and less harsh.  Harsh strings I find jarring and a turn-off.

Don't know which is which.  I'll go with number two as the uncompressed.  So, which is which recording?

???

Computer with upgraded power supply > ASUS Xonar Essence STX > HD-650.

Finally, a Head-Fi thread that saves me money! I could hear slight differences, but honestly could not say which was better. I guess I should reconsider my various upgrade plans, buying flacs instead of mp3s, etc. I was happy to read that codecs have improved significantly in recent years. Overall, a very interesting thread that I enjoyed reading very much, thanks to all that contributed.

I'll keep at it though, maybe my 55-year old ears will "discover" the clues after a while. I will make up a few test wavs with some of my favorite test tracks. Having been raised on cassette tapes and being somewhat new to quality audio (or modest quality in my case), I have to say that modern audio gear is just fantastic.

setup for testing:

ubuntu linux > banshee > udac > bravo v2 > Beyerdynamic DT770/80 and Grado SR80

Quote:

Originally Posted by dana789 

Finally, a Head-Fi thread that saves me money! I could hear slight differences, but honestly could not say which was better. I guess I should reconsider my various upgrade plans, buying flacs instead of mp3s, etc. I was happy to read that codecs have improved significantly in recent years. Overall, a very interesting thread that I enjoyed reading very much, thanks to all that contributed.

 

I'll keep at it though, maybe my 55-year old ears will "discover" the clues after a while. I will make up a few test wavs with some of my favorite test tracks. Having been raised on cassette tapes and being somewhat new to quality audio (or modest quality in my case), I have to say that modern audio gear is just fantastic.

setup for testing:

ubuntu linux > banshee > udac > bravo v2 > Beyerdynamic DT770/80 and Grado SR80

These things have to be evaluated on a case by case basis,  I failed to DBT the 128 vs the lossless and almost of the time I fail mp3 vs lossless which is why I happily rip all my CDs to mp3 (VBR 09 lame 3.98)  , I retain the CDs, however I was able to 17/20 DBT a VBR 0 vs the lossless original on one specific track due to the track being recorded very hot and the encoding adding distortion. The alogrithms though much improved are not yet psychoacoustically perfect and are not designed for exceptional cases. That said on the test track provided here the 128 encoding was darned good.
 

To my ears, Vorbis is superior to mp3. I usually can't ABX Vorbis vs wave at these settings.

It would be interesting to know what kind of encoding was done on this file. 128 kbps is completely obsolete. It has been replaced with -V5, which is a VBR setting supposed to average around 128 kbps, but that rather go near 135-140 kbps in practice. Modern encoders, like AAC or Vorbis, are natively VBR.

OK, I cheated. Following an earlier post, I looked at the spectrogram which makes it obvious which is which. I guess I can't hear much above 16khz, as noted above there's lots of energy above 16 khz in the uncompressed file but none in the compressed file. I can't hear the difference.

I use Ubuntu Linux, I found a spectrum analyzer program called baudline (www.baudline.com, Linux only unfortunately) . Just download, follow the directions, and it works. It reads lots of different file formats and works in real-time. very handy, I'm sure I'll be having fun with it. I suspect there's a Windoze and Mac equivalent.

For a recommended track, I like to test stuff with acoustic instrumental music. Would a commercial track be appropriate or might we run afoul of copyright, etc?

Quote:

Originally Posted by chengbin 

 

The encoding settings were V5, and q 0.

 

Isn't the q 0 switch obsolete ? I know that the V settings are completely optimized.

Quote:

Originally Posted by chengbin 

 

Would people be interested if I set up another test, where I remove the 16KHz low passing, increase bitrate fluctuation, and choose another song?

 

If you remove the lowpass, a lot of bits will be allocated to encode the high frequencies, and the already average quality of medium frequencies should become quite worse.

I did a test of lame 3.97 using V5 option versus V5 q0 option. The Head Fi test flacs took 395 seconds to compress to an mp3 with V5 , 1035 sec with V5 q0. File sizes did not change. So I guess if you're not in a hurry, maybe  q0 could give a noticeable improvement? Could that account for the perceived quality of the compressed example in this thread?

I suggest using the Head Fi test disk available on HDtracks.com as a flac.

http://www.head-fi.org/forum/thread/476089/head-fi-and-hdtracks-david-chesky-collaborate-to-assemble-a-headphone-system-test-album

I suspect we would be violating copyright if we posted compressed tracks though. Does anyone know of "open source" music?

Go ahead and post 30 seconds of it.  It's fair use.

I thought of trying the same thing with some of the files available from metaxas.com.  They are close to the best recordings I've ever heard, if not the best, despite the hiss (which shows, among other things, they haven't been post-processed at all after recording).  They are very likely to suffer from sounding flat and unnatural though after compression of any kind.

I did also guess correctly that the files were VBR, which is why they are so good. I'm willing to bet with CBR that they would have been very obviously different even with much cheaper gear.  When I have some distraction-free time I might try testing whether I can consistently hear the differences on lesser gear with compressed tracks.  It's rather fascinating.

Quote:

Originally Posted by dana789 

I did a test of lame 3.97 using V5 option versus V5 q0 option. The Head Fi test flacs took 395 seconds to compress to an mp3 with V5 , 1035 sec with V5 q0. File sizes did not change. So I guess if you're not in a hurry, maybe  q0 could give a noticeable improvement? Could that account for the perceived quality of the compressed example in this thread?

That's very unsure. I looked for information and found this thread : http://www.hydrogenaudio.org/forums/index.php?showtopic=80388

I didn't understand everything. It seems that between 2004 and 2006, -q0 resulted in distorded audio (see post 5). According to post 3, there should be no difference between q3 and q0 in new VBR, but I don't know from which version of Lame the --vbr-new code was defaulted for the -V settings. Before the latest versions, you had to specify it manually.

These matters are so complicated that the only thing to do is using the lastest stable version and stick to the -V settings.

Does the track you're listening to and what you used to rip it have A LOT of effect on it? I will admit, I did terrible with the provided track. I only guessed 5/13, but when I did it with a track I was accustomed to listening to, it jumped to 9/13. One thing I will say, is that I listened mightly with both, and in casual listening, I probably wouldn't be able to tell the difference. 

Quote:

Originally Posted by chengbin 

It was V5, but checking the bitrate through mediainfo showed the average bitrate to be 129Kbps, close enough. 

 

The encoding settings were V5, and q 0. I probably should have eliminated the >16KHz low pass, but I didn't because I thought there was no content above 10KHz on that track. Obviously I was wrong.

 

Would people be interested if I set up another test, where I remove the 16KHz low passing, increase bitrate fluctuation, and choose another song? If you do, please recommend a song for the next test. Thanks.

I think better yet would be posting a v0 or v2 since there's really no need to compress all the way down to v5. With those settings, I doubt anyone will pass the DBT except in the case of a killer sample. Those also naturally have higher low pass settings, around 19.5 khz for v0, so that would be corrected. And as mentioned above, I would just stick to the V setting and not try to micromanage LAME.

Quote:

Originally Posted by AtomikPi 

I think better yet would be posting a v0 or v2 since there's really no need to compress all the way down to v5. With those settings, I doubt anyone will pass the DBT except in the case of a killer sample. Those also naturally have higher low pass settings, around 19.5 khz for v0, so that would be corrected. And as mentioned above, I would just stick to the V setting and not try to micromanage LAME.

Using V5 instead of V0 at least gives us a fighting chance to possibly maybe hear a difference.

The test wouldn't be nearly as interesting at V0 when nobody is able to hear a difference.  At least with the V5 sample we've got some discussion going on about what some people are hearing as possible differences and some ABX results.

Quote:

Originally Posted by tmars78 

Does the track you're listening to and what you used to rip it have A LOT of effect on it?

The ripper, no, exept if you include the encoder in it. Then yes, there are big quality differences between the mp3 encoders.

The track also is very important. The same kind of encoding can lead to transparent results of a given track, and unlistenable results on another.

Depeche Mode - Something to do is a track that I found very easy to ABX.