Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[bblue]

I finally caught it in the act!

When converting DSF to FLAC, some DSF files produce corrupted FLAC files.

1. Some DSF files produce corrupted FLAC's ONLY if the DSP chain is enabled for the conversion.  If the DSP chain is off, the FLAC is fine, but at 352.8k.  

2. Many DSF files produce proper FLAC's whether or not the DSP chain is enabled.  Those that do, always do, those that don't, always don't.

3. A corrupted output FLAC will be many times the size it should be, typically 1-3 GB and I've seen as high as 6.21 GB.  It will play but stutter and skip a lot on any player.  The stuttering and skipping is identical to what I described in an earlier thread describing library FLAC files that had become corrupted since being placed in the library.

4. The source of these DSF files varies.  SACD rips (PS3 style) that I do are usually fine, but with the last two I purchased (a Time-Life Legends series) every track produces corrupt FLAC files in MC.  Native DSF files from Blue Coast Records often produce good FLAC's without DSP enabled, but some produce bad FLAC's only when DSP is enabled.  Other native or ripped sources produce good FLAC's always.

I can provide some DSF source to demonstrate the problem if someone could investigate.

This also explains the slowness I mentioned in the 154 thread between the conversion of one file to the beginning of the next.  It can be up to 5-6 minutes depending on the size of the FLAC or temp file it is writing.  When a corrupt FLAC is not being produced, the conversions go quickly.

--Bill

Oh, and ALL the DSF files convert properly in Korg's Audiogate software.

[Matt]

When converting DSF to FLAC, some DSF files produce corrupted FLAC files.

1. Some DSF files produce corrupted FLAC's ONLY if the DSP chain is enabled for the conversion.  If the DSP chain is off, the FLAC is fine, but at 352.8k.  

I think this may be fixed next build.  The issue was that DSF conversion produces 64-bit files since that's the native rate out of the DSF.  But the conversion needs to use less than 64 bits and wasn't.

[Matt]

The build is public now.  Feedback appreciated.

[bblue]

The build is public now.  Feedback appreciated.

Yeah, I have been testing it.  So far as far as my issues are concerned not a thing has changed. 

In this last round of testing I've noticed a few other clues.

1. Smaller DSF files (running time less than 3:00) are more likely to succeed at 176k.  Some of those will fail at 176k but succeed at 352.8k (no DSP processing enabled).  Larger files will fail at any output speed.  The larger they are the more outrageous the corrupt file sizes are... multiple gigabytes.

2. Often, the first file you try to convert from DSF to FLAC, if it is one that would normally convert successfully (usually) nothing is written to the designated output directory.  Not even the 'Temporary Convert File - xxxx.flac'.  It never shows up.

3. Some times (I'm not sure just what sets it off) after several conversions (successful or not) all output to the designated directory will just stop.  The conversion appears to go through the motions (successful or not), but nothing shows up.  Sometimes, if you close MC and server then restart it it will start working again.

4. Some times when writing what will become a corrupt file, it produces the Temporary Convert File, but leaves it.  Doesn't complete the process.

I hadn't seen #3 or #4 until this version #155, but I had seen #2 a couple of times when it was the first file being converted in #154.  I could never duplicate it then. It just happened once in a while.

I have a file for you to download from my dropbox.  It is a 7z compressed zip file 480MB in size which contain 5 DSD tracks, all of which are problematic.  If you'd rather download them uncompressed individually, those are there too.  Please use these to identify the problem.  I'll send more details in PM.

--Bill

[Matt]

Thanks for the samples.

They all convert correctly for me if I set DSP Studio to resample.  I did 44,100 Hz as a test.

[bblue]

Thanks for the samples.

They all convert correctly for me if I set DSP Studio to resample.  I did 44,100 Hz as a test.

Apples to apples.  Try 176k and 352k.

[bblue]

Matt,

I found that the filesystem manager I use to duplicate files and folders (StableBit DrivePool, sort of a mirroring of files) was causing the problem with no writes to the defined temp directory used as my target for MC conversions.  There's an interaction between it and MC during MC's conversion process when it goes crazy with oversized corrupted files that it holds the temp file(s) in a locked state for far longer than DrivePool believes it should.  The temp file becomes locked in DrivePool as well, which then blocks MC further use of the directory as a target.

But that doesn't seem to have anything to do with the generation of corrupt files.  I moved MC's output directory for transfers to a single standalone drive, not part of the StableBit pool and then repeated tests.  Starting at 44k output on one of the files I sent you "Rolling" I produced good conversions for 44k, 48k, 96k and 176k.  That was a first for that track.

I then tried one of the Time-Life albums 17 tracks at 176k.  All but one of them were corrupted and way huge. The one that converted correctly was the smallest of the group.  Two of those were among the ones I sent you "Buffalo" and "Rascals".  Try those at 176k and see what happens.  One that does seem to cause MC special heartburn is "Moody" which I just added to the account.  If you need any others I'll be happy to add them.

--Bill

[Matt]

I converted all the files at 176,400 Hz as well.  No problems.

[bblue]

How the heck is that possible??

Any suggestions?

[glynor]

How the heck is that possible??

Any suggestions?

Have you tried taking the pooled drives completely out of the equation, as a test?

Copy the source files to a local, regular disk as well.  Make a BRAND NEW folder on the regular disk, and convert to that.

[bblue]

Have you tried taking the pooled drives completely out of the equation, as a test?

Copy the source files to a local, regular disk as well.  Make a BRAND NEW folder on the regular disk, and convert to that.

Yes, in fact.  This was one of the steps done a couple of days prior.  It didn't make any difference in the file corruption problem.

What I find particularly interesting in light of the 156 release, is that as of my installation of 156 (no system changes, no reboot, nothing except the upgrade of MC) the conversions now work perfectly!  Every track, every album, perfect behavior using exactly the same steps and files I had been using all along.

And, Matt didn't reply to me saying that DSD conversions to FLAC 176k worked fine there on all the test files I had sent him until *after* 156 had been announced. Coincidence?

However it happened, I sincerely thank JRiver for fixing this issue.

--Bill