Skip to main content

Topic: Thinking of switching codec - need some advice please (Read 2933 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • halb27
  • [*][*][*][*][*]
  • Developer
Re: Thinking of switching codec - need some advice please
Reply #25
@Lee James:
a) For lame3995o use the link I gave in reply #6. I just rechecked, it works. (In the lame3995o thread the original link doesn't work any more, but I gave the working reply #6 link later).
b) Please share with us a short (max. 30 sec.) snippet of your track. It would be great if you could re-ABX exactly this snippet and provide us with the information about the spot where you find the deviation the easiest to detect.
c) Most of the things you describe when ABXing is what is usually happening when doing a hard ABX test. The details of the difference you describe are in line with bennetng's suspicion. But I think we can make things clearer when we have a track snippet you can ABX.
d) The fact that you might have kind of a hearing disability does not mean that you're not a good candidate for hearing differences. I remember an early mp3 test performed by the german paper c't where very high bitrate mp3 was best detected by a man with a specific hearing disability.
  • Last Edit: 18 June, 2017, 03:00:46 PM by halb27
lame3995o -Q1.7
opus --bitrate 140

  • Lee James
  • [*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #26
Hi, since TOS8 requires a double-blind test and a sample audio file so that others can examine it, can you upload a lossless snippet of this song for us to examine?
Sure, I've attached the first 30 seconds of both the WAV and MP3. I truncated the WAV using audacity, and the MP3 using the lossless mp3CirectCut utility.

Though please note that there is nothing remarkable about this song at all in my opinion. It was just a random track! :)

Just grabbed the aac and vorbis files from Youtube and got some interesting track peaks (see attached screenshot). Just wanted to make sure it is not the cause of audible differences to avoid some useless or even harmful codec tweaking for developers.
I honestly don't understand why you've mentioned YouTube, but YouTube really should not be used for anything to do with audio. YouTube totally destroys audio and adds noise to tracks, similar to a lot of recording devices.

a) For lame3995o use the link I gave in reply #6. I just rechecked, it works. (In the lame3995o thread the original link doesn't work any more, but I gave the working reply #6 link later).
OK thanks, I will try it when I get time. Probably won't be till later in the week. (Lately I've barely had a moment to myself alone with the computer.)

It would be great if you could re-ABX exactly this snippet and provide us with the information about the spot where you find the deviation the easiest to detect.
I will have a think about this, but I really don't think there is any 'spot' which is easiest to detect - it all sounds much the same to me.
  • Last Edit: 18 June, 2017, 04:10:10 PM by Lee James

  • Lee James
  • [*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #27
If there is anything 'wrong' with the files I've been using, I hope someone can figure out what. I did think that the two tracks had the same gain but maybe I was wrong? It would be nice if I was somehow 'wrong' about all this, though I'm not sure how I could be, since I have simply encoded the files using the LAME source (V0) and nothing more complicated than that, so the mp3s should be the same gain right? I think I probably added ReplayGain tags, but ABX obviously ignores that.

  • bennetng
  • [*][*][*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #28
Sure, I've attached the first 30 seconds of both the WAV and MP3. I truncated the WAV using audacity, and the MP3 using the lossless mp3CirectCut utility.
If you are going to redo the test, you should encode the trimmed wav file to mp3 instead of encode the whole song to mp3 then cut it, since the mp3 file you attached is almost one second longer than the wav.

Quote
Though please note that there is nothing remarkable about this song at all in my opinion. It was just a random track! :)
Does it mean you have a good chance of telling mp3 V0 and lossless apart, no matter what file it is? Then I am more interested to know if you strictly need lossless, or another supposedly better lossy codec may suffice. In this case I would provide several lossless and correctly encoded (time aligned, no gain problem) lossy files in different genres for you to try so you don't need to worry about doing something technically wrong. Are you interested?

Quote
I honestly don't understand why you've mentioned YouTube
Since you didn't provide a sample in your previous reply. With a lossless sample Youtube reference is not useful anymore.

If there is anything 'wrong' with the files I've been using, I hope someone can figure out what. I did think that the two tracks had the same gain but maybe I was wrong? It would be nice if I was somehow 'wrong' about all this, though I'm not sure how I could be, since I have simply encoded the files using the LAME source (V0) and nothing more complicated than that, so the mp3s should be the same gain right? I think I probably added ReplayGain tags, but ABX obviously ignores that.
It is a Windows issue, not specific to playback software or replaygain.
https://hydrogenaud.io/index.php/topic,104051.0.html

  • Lee James
  • [*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #29
If you are going to redo the test, you should encode the trimmed wav file to mp3 instead of encode the whole song to mp3 then cut it, since the mp3 file you attached is almost one second longer than the wav.
Oh, that's odd. I typed in 30 seconds for both.

Does it mean you have a good chance of telling mp3 V0 and lossless apart, no matter what file it is? Then I am more interested to know if you strictly need lossless, or another supposedly better lossy codec may suffice. In this case I would provide several lossless and correctly encoded (time aligned, no gain problem) lossy files in different genres for you to try so you don't need to worry about doing something technically wrong. Are you interested?
Sure, I'd be interested in any kind of tests you have, though I don't have a whole lot of time at the minute so can't promise to do a lot of testing.

All I really started this thread for is I just want a really good codec that's transparent for me - having done my testing on LAME, I am generally reassured about it, though if there are imperfections then I will consider a better codec? :) As I said from the start, I'm not someone who particularly wants or needs lossless as I like the idea of stripping away unneeded information and having smaller files.

It is a Windows issue, not specific to playback software or replaygain.
https://hydrogenaud.io/index.php/topic,104051.0.html
That thread is 'over my head'. In simple terms, do you think this could be just a simple matter of the gain? If that were the case then I would be delighted. I don't really care if there's some bug that makes my mp3s slightly quieter than the original, just as long as they're transparent when played at the same volume.

  • bennetng
  • [*][*][*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #30
Sure, I'd be interested in any kind of tests you have, though I don't have a whole lot of time at the minute so can't promise to do a lot of testing.
Don't worry. It is unfair to push you to do some quick test in a very short time. I will allow about 3 months or so for you to do about 5 files of different genres, and encode them with different format/settings.

Quote
That thread is 'over my head'. In simple terms, do you think this could be just a simple matter of the gain? If that were the case then I would be delighted. I don't really care if there's some bug that makes my mp3s slightly quieter than the original, just as long as they're transparent when played at the same volume.
As shown in the video, all files are played with RG disabled and identical playback volume. The wav file nulled to 24-bit noise floor in the WASAPI vs DS test, it means the peak level of the wav file is low enough to disable the DS limiter, unlike the mp3 file.

https://youtu.be/v26f0gfNzXI

The nulled 24-bit flac files are attached to indicate Windows mixer obviously limited the peak area of the mp3 file. The post below explained the nature of null tests.
https://hydrogenaud.io/index.php/topic,105377.0.html

  • shadowking
  • [*][*][*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #31
With all due respect, a -single- trial of 12/16 may not be very significant unless your confident of the artifact. Unfortunately with such difficult tests you got to allow more trials if you predict less than perfect scores , or get a better one on a single trial like 8/8, 14/16 etc
wavpack 4.8 -b4x4s0.75c

  • halb27
  • [*][*][*][*][*]
  • Developer
Re: Thinking of switching codec - need some advice please
Reply #32
@Lee James:
I've looked up your files, and as the peak values of the mp3 file are above 1.0 bennetng's remarks may be the reason for your experienced differences.
So in order to make sure that it isn't Windows' machinery which drops in and makes the mp3 track quieter (which could lead to the audible differences you describe) my suggestion is:
a) convert your 30 sec. wav snippet to flac
b) re-encode the flac file
c) let foobar compute the replaygain values of the flac and the newly created mp3 file
d) ABX the flac and the newly created mp3 with the Replaygain checkbox on.

The track has a replaygain value of ~-7.75 db, so when using Replaygain while ABXing Windows does not make the mp3 track quieter. This makes up for a fair comparison.
lame3995o -Q1.7
opus --bitrate 140

  • Lee James
  • [*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #33
Bennetng >> thanks for the video and the info. I'm afraid this is still all 'over my head' and a little too technical for me to understand, sorry. You're saying that the mp3 will play quieter I guess.

Halb >> OK, sure, I will do that. I assume that step (b) is encoding to mp3 (rather than re-encoding to FLAC). If I do this, will it definitely be a fair test in your opinion?

  • halb27
  • [*][*][*][*][*]
  • Developer
Re: Thinking of switching codec - need some advice please
Reply #34
No, you need a lossless format that can store the replaygain value. You can't do that with a wav file. So a procedure like the one I mentioned is necessary.
lame3995o -Q1.7
opus --bitrate 140

  • bennetng
  • [*][*][*][*][*]
Re: Thinking of switching codec - need some advice please
Reply #35
What halb27 said.  You should also upload the newly created flac and mp3 file so that we can examine them. If you are going to participate in some further listening tests I mentioned in my previous post, compressed lossless files will also be used.

  • Vivadavid
  • [*]
Re: Thinking of switching codec - need some advice please
Reply #36
Thank you! In that case, I'll consider that there's transparency if I don't get more than 5% in my ABX.

In the Wiki Article an error of only 1 percent is allowed whereas the foobar ABX procedure allows for 5 percent.
Being within a guessing probability of 5 percent maximum is what is usually accepted here.