Flac discrepancies

[kevmitch]

I have found that the directors cut on my Aliens bluray (part of the 4 volume anthology) causes makemkv to produce flac output that is inconsistent with eac3to and mmdtsdec. In addition, the makemkv waveform "looks wrong" at the point of discrepancy. For the makemkv flac, the procedure is

• Rip title with the flac profile
• Extract the flac to the separate file makemkvflac.flac with mkvextract

I compared this to the result of

• Rip title as is with default profile
• Use eac3to to extract and decode the dtshd track to dtshd.eac3to.flac

The files start out identical, but staring at 509.845s (24472576 samples) they differ. This is clearly visible in this plot. It is also pretty clear from the abrupt jump that it is the makemkv version that is not correct. The flac files also differ in their number of samples (dtshd.ea3to.flac=444814848 vs. makemkvflac.flac=444820992).

For further reference, I also tried to

• Rip title with default profile
• extract dtshd track to dtshd.dtshd with mkvextract
• convert the dtshd to wav with mmdtsdec

This produced the error output below

Code: Select all

$ mmdtsdec.exe -d 'C:\Program Files (x86)\eac3to\dtsdecoderdll.dll' dtshd.dtshd dtshd.mmdtsdec.wav
DEBUG: code=0001dd3c src=^1Qd^s!Q9l)/c7+Fhb:121261681
DEBUG: code=00000000 src=8]R6+RR hnJnFM7k:121263912
Error: Failed to decode

The refernece flac encoder (and my comparison script) was not able to read the resulting .wav file. However, I was able to use eac3to to convert it to a functioning flac. The result was identical to the output of the direct eac3to decode of the dtshd track.

It therefore appears that the problem is specific to the way the makemkv creates the flac track on the fly.

[mike admin]

This is a bug Please share the following:
- full MakeMKV conversion log, when using flac profile (i.e. log file written by MakeMKV, should contain additional information).
- the extracted dtshd.dtshd from default profile.

Also, I would appreciate if you can run the test (using flac on the fly conversion) one more time and check that the problem is persistent - MakeMKV should produce the same flac track. If it produces a different flac track (same as reference eac3to or with the difference at another place) then please mention this as well.

Thank you.

[kevmitch]

This is a bug Please share the following:
- full MakeMKV conversion log, when using flac profile (i.e. log file written by MakeMKV, should contain additional information).

The conversion log

- the extracted dtshd.dtshd from default profile.

The resulting file is ~4GB which is larger than dropbox will let me upload. What is the best alternative?

Also, I would appreciate if you can run the test (using flac on the fly conversion) one more time and check that the problem is persistent - MakeMKV should produce the same flac track. If it produces a different flac track (same as reference eac3to or with the difference at another place) then please mention this as well.

Tried two more times and makemkv produces identical tracks each time with an identical error.

[kevmitch]

The same thing happens with the theatrical cut on the same disc in the same place. As usual, the theatrical and directors cut are just different assemblies of the same pool of segments. Both the directors cut and the theatrical release have the same first segment which is 8:31 which puts the error most likely after the segment boundary. Unfortunately, even the first segment is over 2GB. Instead, I've uploaded the full backup skeleton with all the streams removed. Included instead are the .dtshd tracks from the first two segments of the directors cut as demuxed by tsmuxer.

[mike admin]

Please collect a testdump as well per instructions at /onlinehelp/testdump.html .

[kevmitch]

The testdump [link downloaded]

[kevmitch]

I got my hopes up reading about "seamless sample level joining" in the 1.8.8 changelog. Unfortunately, I'm still seeing essentially the same thing with this new version. I say "essentially" because it appears that the first sample where eac3to and makemkv differ is slightly different with 1.8.8: 24472448 vs 1.8.7:24472576.

[kevmitch]

I remember mikeadmin mentioning that mmdtsdec could talk to a tcp port instead of writing a broken .wav file. Does anyone know how that works (i.e., how to set up the port and how the data is encoded), or at least where to look in makemiv-oss to see where this is going on. I did a quick grep for "dts" and "mmdtsdec",but didn't find anything. Maybe all the magic is going on behind closed doors in makemkvcon?

Until this is on-the-fly flac encoding works losslessly, this would save me having to fart around with eac3to.