Noise between two VBR Files (Lame 3.98.2) Options

[axiom]

I wasn't able to find anything about this in here, so I decided to register.

I encountered a strange effect today: Having perfect gapless playback between two FLAC tracks, I converted them to MP3 using LAME 3.98.2 (--noreplaygain -V 0 --vbr-new). Playing the files with normal volume created the illusion of gapless playback, but when I raised the volume (disco level) I was able to hear a significant noise between the two tracks. Using --vbr-old produced a clear noise too, but not so loud like the one with --vbr-new.

Just to make it more confusing, converting it with cbr settings produced gapless tracks (--noreplaygain -h -b 320).
And a transition between a VBR to a CBR (320 / 256) or FLAC file was gapless too
So the noise is at the beginning of the 2nd VBR file. And just for fun, encoding it with 192kb/s in CBR mode produced also a noise...

Just to accentuate it, you have play the files very loud. I compared them many times with the FLAC files in the same volume, and the FLAC files would play gaplessly. (With cracy volume settings, you can also hear a very silent noise between the CBR files, but never mind, I don't want to become deaf so early )

My results (transition between two files):

flac -> no noise
cbr 320 / 256 -> no noise
cbr 192 -> noise at the beginning of the 2nd track
vbr new v0 / v2 -> noise at the beginning of the 2nd track
vbr old v0 / v2 -> noise at the beginning of the 2nd track, but not as loud as vbr-new or cbr 192



Source:

######################################################################

Exact Audio Copy V0.99 prebeta 5 from 4. May 2009

EAC extraction logfile from 29. June 2009, 1:59

The Phantom of the Opera / The Phantom of the Opera (OST)

Used drive : TOSHIBA DVD-ROM SD-M1612 Adapter: 1 ID: 1

Read mode : Secure
Utilize accurate stream : Yes
Defeat audio cache : No
Make use of C2 pointers : Yes

Read offset correction : -472
Overread into Lead-In and Lead-Out : No
Fill up missing offset samples with silence : Yes
Delete leading and trailing silent blocks : No
Null samples used in CRC calculations : No
Used interface : Installed external ASPI interface
Gap handling : Not detected, thus appended to previous track

Used output format : User Defined Encoder
Selected bitrate : 768 kBit/s
Quality : High
Add ID3 tag : No
Command line compressor : D:\Programs\Exact Audio Copy\FLAC\FLAC.EXE
Additional command line options : -8 -T "ARTIST=%a" -T "TITLE=%t" -T "ALBUM=%g" -T "DATE=%y" -T "TRACKNUMBER=%n" -T "GENRE=%m" %s -o %d


TOC of the extracted CD

Track | Start | Length | Start sector | End sector
---------------------------------------------------------
1 | 0:00.00 | 2:47.05 | 0 | 12529
2 | 2:47.05 | 3:40.62 | 12530 | 29091
3 | 6:27.67 | 2:43.50 | 29092 | 41366
4 | 9:11.42 | 1:59.29 | 41367 | 50320
5 | 11:10.71 | 3:36.10 | 50321 | 66530
6 | 14:47.06 | 5:42.02 | 66531 | 92182
7 | 20:29.08 | 3:29.55 | 92183 | 107912
8 | 23:58.63 | 4:54.12 | 107913 | 129974
9 | 28:53.00 | 2:16.34 | 129975 | 140208
10 | 31:09.34 | 5:31.40 | 140209 | 165073
11 | 36:40.74 | 3:42.71 | 165074 | 181794
12 | 40:23.70 | 8:00.72 | 181795 | 217866
13 | 48:24.67 | 12:48.48 | 217867 | 275514
14 | 61:13.40 | 2:21.35 | 275515 | 286124


Track 4

Filename D:\The Phantom of the Opera\The Phantom of the Opera (OST)\04 - The Mirror (Angel of Music).wav

Peak level 56.0 %
Track quality 100.0 %
Copy CRC AA0AEDB4
Accurately ripped (confidence 9) [2EECE574]
Copy OK

Track 5

Filename D:\The Phantom of the Opera\The Phantom of the Opera (OST)\05 - The Phantom of the Opera.wav

Peak level 92.5 %
Track quality 100.0 %
Copy CRC 550E91B0
Accurately ripped (confidence 8) [D0319F42]
Copy OK


All tracks accurately ripped

No errors occurred

End of status report

######################################################################


Player:
Foobar2000 v0.9.6.3


Anyone else?

This post has been edited by axiom: Jun 29 2009, 02:06

[pdq]

You left out one key piece of information...how are you playing back the files?

MP3 is not inherently gapless so you need the right combination of encoding and playback to make it so. How you ripped the files has no bearing.

[axiom]

You left out one key piece of information...how are you playing back the files?

MP3 is not inherently gapless so you need the right combination of encoding and playback to make it so. How you ripped the files has no bearing.

sry

I'm using fb2k v0.9.6.3

[[JAZ]]

You could do one thing to help on detecting this:

With foobar, select the two songs (in mp3 vbr), right click, go to convert, select the "merge all tracks into one output file" and decode to .wav.

Then using audacity or any other audio editor, cut ten seconds, where the second 5 would be precisely the change from one song to the other.

Finally, encode this to flac and upload it to the forums (There is an upload forum here where you can upload some files) and add a link from this post to that post.


Another different thing you could try is using mp3directcut, cut the last seconds of the first mp3 and the first seconds of the second mp3 and upload both fragments. It may not maintain sample accuracy in this way, so it's a second best.

[pdq]

Another different thing you could try is using mp3directcut, cut the last seconds of the first mp3 and the first seconds of the second mp3 and upload both fragments. It may not maintain sample accuracy in this way, so it's a second best.

This would remove the information that fb2k uses to produce gapless playback, so I don't see how this is going to help.

[rpp3po]

For continued research without hearing damage use Synchrotron. Convert the 2nd VBR file to WAV with Foobar's diskwriter. Then let Synchrotron calculate its delay vs. the original. It should be zero! Else either Foobar's gapless implementation is flawed or LAME is adding invalid info. It could be, though, that there actually isn't any delay, but that LAME is producing an artifact at the beginning of the file.

This post has been edited by rpp3po: Jun 29 2009, 12:54

[[JAZ]]

This would remove the information that fb2k uses to produce gapless playback, so I don't see how this is going to help.

That information can be put back (with foobar itself). Also, since it is not a cut inbetween but instead, from the start, the program may even keep that information (I just haven't verified if it is possible). Anyway, "second-best" meant really that.

[axiom]

Samples


I also did the delay checks with Synchroton, but no delay was found

Ouput:

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"cbr 192 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.9994424110474592

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"cbr 256 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.99983952789585

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"cbr 320 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.999941835369027

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"vbr-new V0 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.9994572740376045

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"vbr-new V2 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.9990205462665336

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"vbr-old V0 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.9996800981847263

D:\Synchroton>java -jar Synchrotron.jar "flac 05 - The Phantom of the Opera.wav"
"vbr-old V2 05 - The Phantom of the Opera.wav"

PCM_SIGNED 44100.0 Hz, 16 bit, stereo, 4 bytes/frame, little-endian
Delay: 0 Cross Correlation: 0.9994059028051483

[greynol]

I presume this noise that to which you're referring is a click?

Remember, we're talking about lossy encoding here. You cannot guarantee that there won't be a discontinuity in amplitude between the junction of two individual files.

http://www.hydrogenaudio.org/forums/index....mp;#entry632617
http://www.hydrogenaudio.org/forums/index....mp;#entry624299

This post has been edited by greynol: Jun 29 2009, 22:16

[axiom]

Yes, it's a click, I messed it up a little with partially talking about gapless or not.

[greynol]

Had you characterized the nature of the "noise" in the first place, we could have addressed your problem yesterday.

I doubt I'm alone in originally thinking you were talking about increased noise, as in white* noise.

(*) or any other color of noise for that matter.

This post has been edited by greynol: Jun 29 2009, 22:20

[axiom]

So it's maybe a "known-issue"? But why is the result with VBR V0 not the same like the one with CBR 320?

I was very happy to get gapless playback with mp3 working, but your posts disillusion me. Should I just encode the critical files with CBR 320 or better use FLAC when a transition between two tracks is present?

[greynol]

Of course it's a known issue.

Furthermore, it should not be surprising that different encoders and encoding settings will give varying degrees of the problem.

If this is that major of an issue for you and you must use a lossy format, perhaps you should consider combining the lossless tracks and then encoding. I'm not 100% sure if this works, but maybe you can then split it with pcutmp3.

This post has been edited by greynol: Jun 29 2009, 23:10