Print

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

[TheLion]

I am aware that I open a can of worms with this very topic. But after spending hours and hours comparing different options for audio decoding I have convinced myself (approved by my wife in double blind tests ;-)) that different decoders deliver (slightly) different sound even with the exact same source.

I come from archiving all my Blu-Rays with eac3to to mkv. All audio streams are transcoded to flac. And here we go - I have changed my opinion a few times about it, couldn't believe it at first BUT now I am ready to say something stupid in public: Decoding those flac streams with madFlac or ffdshow (32bit fp processing/output) delivers (slightly) different sound. MadFlac being "fuller", a touch less "aggressive" in the highs. "Smoother" if you will (and now I am out of audiophile vocabulary...).

Next stop: ripping Blu-Rays 1:1 with AnyDVD. Decoding the various audio streams of the resulting bdmv with LAVaudio. I did mostz testing with DTS HD MA tracks using Arcsoft decoder 1.1.0.8 (both times with eac3to and LAVaudio). Therefor I am comparing a Blu-Ray ripped with eac3to/Arcsoft to mkv/flac with the original bdmv decoded by LAVaudio/Arcsoft.

In general the results given we are talking about lossless audio should be exactly the same. Well - during playback they aren't. The mkv/flac decoded by madFlac is again "fuller", "smoother". The LAVaudio output is "brigther", almost annoying in the highs.

Am I crazy? And my wife with me? Am I imagining things even in double blind tests?

One word about my setup - it is probably as revealing as they come. I connect per ASIO to a Prism Sound Orpheus Firewire Interface (in async mode, in slave mode the difference is almost the same as with madFlac versus LAVaudio/ffdshow -> the sound is fuller/smoother against brighter/more aggressive). Than per XLRs (I am not mentioning my voodoo cables ;-) - ok, they are Vovox) into Parasound JC-1 monoblocks. Speakers are Danley SH-50.  

[jmone]

So if I read this correctly your two playback chains are:

1) DTS-MA --> eac3to/arcsoft --> PCM --> madflac --> flac (in MKV) --> madflac --> PCM --> AV equipment
2) DTS-MA --> lav/arcsoft --> PCM --> AV equipment

It would be interesting to compare the three different PCM streams to see if they are bit identical ....but from your listening I'd say that at least one is different and hence something is not a truly lossless process.  Assuming that the initial decoding of the DTS-MA are both being done by the same version of arcsoft's decoder it stands to reason any difference could be either:
a) The configuration settings from eac3to / lav on the arcsoft decoder are different (eg the arcsoft decoder has various "effects and mixing" that may or may not being used) OR
b) The additional flac encoding/decoding component is changing the output

Only madshi / nevcairiel really know how they call the arcsoft decoder....

EDIT:  To help narrow it down, if you use eac3to to create a MKV with a PCM track (instead of a FLAC), what does it sound like when played?

[sskings]

Agree with the suggestion.  There are a number of free utilities available that compare files to see if they're identical.  Try one on the PCM files.  I've done this successfully before to confirm that transcoders are operating perfectly.

[BryanC]

You should note that Arcsoft decoder 1.1.0.8 has decoding (encoding too?) issues. I believe that 1.1.0.1 is the preferred decoder.

http://forum.doom9.org/showthread.php?p=1508578#post1508578

[TheLion]

Arcsoft doesn't have any decoding "issues" - other than mono to stereo and 6.0 decoding. Which are both not relevant for me.

[BryanC]

Arcsoft doesn't have any decoding "issues" - other than mono to stereo and 6.0 decoding. Which are both not relevant for me.

OK, well I'm not a mind-reader.

[Hendrik]

If you're using the ArcSoft decoder both in eac3to and in LAV Audio, it should indeed sound the same, as its all done in the ArcSoft "black box". Bit-comparing the resulting PCM would be interesting, but i really dont have that many settings that i do pass to the decoder.

I can probably do the bit comparison myself if i one day feel like it. However, which version is "right", i don't know.
You should do what was suggested before, and store a PCM stream out of eac3to in a MKV, to drop the FLAC element, maybe that narrows it down a bit.

[Hendrik]

Ok i did a quick comparison here, using eac3to 3.24 with ArcSoft DTS 1.1.0.8, and LAV Audio 0.29, as well with ArcSoft DTS 1.1.0.8

First Test
------------
Created a raw PCM .wav file from a DTS-HD stream with eac3to (so no FLAC middle-man), and saved the PCM output of LAV Audio decoding the DTS-HD to a file as well.
Comparison: The files are 100% exactly the same.

Second Test
--------------
Created a FLAC file with eac3to, played it with LAV Audio, and compared results to the results from the first test.
Again: The Output is exactly the same as the previous two files.

Third Test
-------------
Decoding the FLAC file from the previous test with madFLAC 1.10
The file is mostly the same. Its basically 100%, but for some reason there is around 10kB of extra data at the end of the PCM output thats not present in any other files. However, all the data before that is 100% exactly the same, so it wouldn't have changed the audio output.
It wasn't even present in the original .wav file created by eac3to, so not sure where its coming from.


So all tests prove two things: Decoding is indeed lossless (of both DTS-HD with ArcSoft and of FLAC), and the output from madFLAC and LAV Audio is basically 100% identical as well.
Maybe something funny is going on with your system. Are you sure eac3to is using ArcSoft (and not just dumping the DTS core)? AFAIK, it requires a bit more setup before it uses it, it only uses it when its properly available over DirectShow, copying in the dtsdecoderdll.dll is NOT enough for eac3to, but it is for LAV Audio.

Maybe its something that only happens on certain files. I only tested one 5.1 DTS-HD MA sample, not any funny setups.

Or maybe, your mind is playing tricks on you.

[TheLion]

Thank you so much for your efforts!

I have no doubt in my mind that both (all correctly designed) solutions output bit-"perfect" and therefor bit-identical streams. Just as any (correctly designed) standalone player should output bit-identical streams on its digital outputs when dealing with lossless (and no other things like mixing introducing changes on a bit level).

Fact is they do sound different (to me). My first thought was - my mind is playing tricks with me. I hear things because I know that these are two "different" options. But double blind test. With my wife and a several friends as well? I am certainly no audiophile who believes in "snake oil". I am a scientist at heart and in profession.

But IMHO bit-identical is just one part of the story. By that logic ANY digital chain should sound pretty much the same. Any media player with "neutral settings" should sound the same with lossless files and digital outputs...

Bits are one part - the other is timing. The time domain. "Jitter" to use this buzz word. My setup is special in that I use a convolution VST plugin which applies FIR filtering. This process is very time domain "sensitive"/critical because the filtering applies to group delay/time domain as well as the frequency response. After the convolution plugin "Jitter" isn't important at all for me. With the Prism Sound Orpheus I use an asynchronous Firewire audio interface with state of the art clocking (it is its own master clock - the PC is just a data pump).

As I said with my initial post - the difference there is between madflac and LAVaudio output (both using Arcsoft and the same input stream) is comparable to putting my audio interface in sync or async mode. In both these cases the exact same bit-identical stream is sent to the audio interface. Very few people would doubt that sync/async modes of audio interfaces sound significantly different. I guess the convolution makes my setup quite jitter sensitive.

I have a couple of issues with LAVaudio/splitter. When playing bdmv's (direct copies of BD's) I see cases of getting severe audio and video async. Sometimes a very strange thing happens and the playback (video and sound) seems in fast forward mode (playing faster than normal). I have to pause playback to fix that.

I am asking myself how LAVaudio/splitter is dealing with seamless branching? eac3to is correcting async problems (delays, breaks) in such branched streams. How is LAV dealing with it?

Thank you! Schöne Grüße nach Deutschland!

[Hendrik]

LAV Audio has a suffisticated jitter correction which causes it to output 100% jitter free audio, unless there indeed is a gap or overlap in the audio.

Regarding seamless branching - of course a "offline" application can do a bit more then a playback component, however i've not had any reports that there are issues with A/V sync in that case, and manual inspection of the timestamps on the file changes did always seem perfect. Note that i can't test this on every disc ever made, so i only did checks on some discs to make sure the code works in theory.

[JohnT]

... But double blind test. With my wife and a several friends as well? I am certainly no audiophile who believes in "snake oil". I am a scientist at heart and in profession...

Did you use an SPL meter to make sure the volume from the speakers was identical between the two sources during the double blind test?  A small difference in volume can make a world of difference in the perception of quality by the listener.

[SteveGoff]

Did you use an SPL meter to make sure the volume from the speakers was identical between the two sources during the double blind test?  A small difference in volume can make a world of difference in the perception of quality by the listener.

But if both files output PCM there should be no difference, and any difference would easily show up in nevcairiel's comparisons.

[Blaine78]

Hi,
like to chime in my 'crazy' thoughts. been using WAV as my format of choice, used to use FLAC. without any doubt, from same source CD, or, converting bit identical FLAC to WAV, WAV wins. sounds a tad more open and less strained in general, and closer to how it sounds originally on CD. This observation, also shared by others on audiophile sites, will get challenged, and i'm very used to the arguments. just my 2 cents

[BryanC]

Hi,
like to chime in my 'crazy' thoughts. been using WAV as my format of choice, used to use FLAC. without any doubt, from same source CD, or, converting bit identical FLAC to WAV, WAV wins. sounds a tad more open and less strained in general, and closer to how it sounds originally on CD. This observation, also shared by others on audiophile sites, will get challenged, and i'm very used to the arguments. just my 2 cents

I have no doubt that you hear a difference between a FLAC and WAV file. I do not believe, however, that there is an actual difference. The placebo effect is very powerful and should not be disregarded.

[mojave]

The placebo effect is very powerful and should not be disregarded.

True, but it also works both ways. Those that don't expect to hear differences might not even if there are differences. 

[jmone]

EDIT:  To help narrow it down, if you use eac3to to create a MKV with a PCM track (instead of a FLAC), what does it sound like when played?

I'd still be interested to hear back from TheLion where is audible difference is introduced....

[BryanC]

True, but it also works both ways. Those that don't expect to hear differences might not even if there are differences. 

True, but mathematics is on my side, and numbers don't lie. The only thing I believe in audio are numbers, since ears are a biological device and thus can change from instant to instant.

If you compress 5MB of pure text into a zip file and then decompress it, if it reads the exact same text, nothing has been lost. Not in theory, in actuality. Same goes for FLAC. A bit is a bit is a bit. If you want to argue that they sound different, then you have to accept that if you make a digital copy of a WAV file, it has the potential to sound different from the file it was copied from. If you agree with this, then there must be a fundamental flaw in digital computing that has yet to be discovered, which has a much larger implication on the world beyond just music.

[Blaine78]

I have no doubt that you hear a difference between a FLAC and WAV file. I do not believe, however, that there is an actual difference. The placebo effect is very powerful and should not be disregarded.

thing is, i don't  have to try and hear it. i hear it, in blind test setup by friend, i hear it.
also, notice how nobody has ever said FLAC sounds better, only ever  WAV, numerous times, by quite a number of people. Also, I was never peer pressured, as I found out for myself this audible difference, then found forums about it. doesn't sound like placebo, infact i want to use flac, better tagging support, and half the size of WAV, i don't inconvenience myself for no good reason. i have 3000 albums in WAV.

[BryanC]

thing is, i don't  have to try and hear it. i hear it, in blind test setup by friend, i hear it.
also, notice how nobody has ever said FLAC sounds better, only ever  WAV, numerous times, by quite a number of people. Also, I was never peer pressured, as I found out for myself this audible difference, then found forums about it. doesn't sound like placebo, infact i want to use flac, better tagging support, and half the size of WAV, i don't inconvenience myself for no good reason. i have 3000 albums in WAV.

Take a WAV file, convert it to FLAC, then convert it back to WAV. Do the original WAV file and the resulting WAV file sound different? Are they bit-identical?

[Blaine78]

Take a WAV file, convert it to FLAC, then convert it back to WAV. Do the original WAV file and the resulting WAV file sound different? Are they bit-identical?

Yes, I've checked this, they are bit identical.
My theory is not with the file itself, but the processing, and path the data takes. Which, does make sense as FLAC needs that little extra CPU to process in realtime. Whether or not the noise created by the extra needed CPU clock cycles be the reason, the audible differences are slight at best, but enough for me to notice with familiar music and a very familiar, and revealing system. I'm not condoning using WAV over FLAC, but the difference I hear is enough for me.

[JimH]

We had a similar discussion today as a result of what another customer believed about the effect of the CPU and other PC activity on playback.  So I did a little testing.  Playing an 88.2 Khz FLAC file took around 1% of my CPU.   I'm not sure a WAV file would take less, but I can't see how it could have any effect on playback.

My PC is an I7 so it's not a slow machine.  The JR Mark was about 3800.

[Blaine78]

We had a similar discussion today as a result of what another customer believed about the effect of the CPU and other PC activity on playback.  So I did a little testing.  Playing an 88.2 Khz FLAC file took around 1% of my CPU.   I'm not sure a WAV file would take less, but I can't see how it could have any effect on playback.

My PC is an I7 so it's not a slow machine.  The JR Mark was about 3800.

i have an i3 at full speed, not a slow machine either. is an odd one. there could be other variables at play here, including just the tiny amount of processing for constant decoding.

[DarkPenguin]

Am I crazy?

Yes.

Start with an spl meter.

[Blaine78]

Yes.

Start with an spl meter.

sound pressure level metre has nothing to do with harmonics and timbre of audio.

[DarkPenguin]

sound pressure level metre has nothing to do with harmonics and timbre of audio.

That's super.  When did I say that it did?

[Blaine78]

That's super.  When did I say that it did?

guess when you said use a SPL meter to someone describing that they hear fuller, smother, or brighter sound, not a louder sound

[craigmcg]

John T and Dark Penguin both suggested the use of a SPL meter. I agree with them because I remember reading in audiophile magazines (before there were sites) that exact gain or volume matching was essential in critical listening. This was because a difference of as little as 1dB typically resulted in the empirically louder (tested by measurement) of the two samples being perceived as sounding better, fuller, brighter, etc. than the other, but not louder. Again, if memory serves me correctly, it typically takes about a 3dB difference for most people to recognize one sample as being louder than the other as opposed to merely being different.
 
My belief is that very slight volume differences between your files are the cause of the fuller, smoother, brighter, but not louder sound you describe.

This said, it's your music and movies and the most important thing is that you enjoy them in whatever formats you prefer.

As for me, as I approach 50, I can't hear the difference between my FLAC and VBR V0 MP3 files anyway (but I still  keep both formats available) and I still haven't figured out Blu-Ray ripping to FLAC within MKV (sigh, maybe someday).

[Blaine78]

John T and Dark Penguin both suggested the use of a SPL meter. I agree with them because I remember reading in audiophile magazines (before there were sites) that exact gain or volume matching was essential in critical listening. This was because a difference of as little as 1dB typically resulted in the empirically louder (tested by measurement) of the two samples being perceived as sounding better, fuller, brighter, etc. than the other, but not louder. Again, if memory serves me correctly, it typically takes about a 3dB difference for most people to recognize one sample as being louder than the other as opposed to merely being different.
 
My belief is that very slight volume differences between your files are the cause of the fuller, smoother, brighter, but not louder sound you describe.

This said, it's your music and movies and the most important thing is that you enjoy them in whatever formats you prefer.

As for me, as I approach 50, I can't hear the difference between my FLAC and VBR V0 MP3 files anyway (but I still  keep both formats available) and I still haven't figured out Blu-Ray ripping to FLAC within MKV (sigh, maybe someday).

not sure i get where you are coming from. Smooth, bright, fuller has nothing to do with SPL, it's perceived loudness that you are thinking of, hence, 'sound pressure level.'

[mojave]

We had a similar discussion today as a result of what another customer believed about the effect of the CPU and other PC activity on playback.  So I did a little testing.  Playing an 88.2 Khz FLAC file took around 1% of my CPU.   I'm not sure a WAV file would take less, but I can't see how it could have any effect on playback.

My PC is an I7 so it's not a slow machine.  The JR Mark was about 3800.

I have an I7 920 running at 2793 Mhz per CPU-Z (yes, I really should overclock). This week I bought and connected a Steinberg MR816. I played some songs in MC16 using ASIO output and was playing files from memory. I listen while I work and every time I opened Internet Explorer or opened a tab I would get a slight pause in the music. I followed the Steinberg troubleshooting recommendations and disabled CPU EIST in the bios. I also set background processes to have priority. I messed with more and less buffers in the drivers and adjusted buffers in MC16. I also checked and unchecked "Use large hardware buffers." Nothing would get rid of the brief pause in music when opening IE. Other browers (Chrome and Firefox) did not affect the music. I even resorted to turning of almost all services using Fidelizer. Still had the brief pause during playback.

I checked my memory and it was running at 533 Mhz (bus clock speed) at 1.5 volts. Per the SPD, my memory can run at 800 Mhz with 1.65 volts. I changed the settings in the bios to increase the memory speed, but left the timings the same (8,8,8,24). This got rid of the brief pause! I have no idea why. I did realize just now that I never tried playback with "play files from memory" unchecked. I took the MR816 home last night so I can't test any more today. I do find it interesting that such a small change affected playback.