New OSX Opensource audiophile player : Audirvana

[Blu]

In my last post I wrote how the 0.8 GCC version was more realistic and involving compared to the official 0.8 version.

 

Well I've been having a listening session and have noticed that the bass frequencies in the new version are very noticeable, I would even go as far as saying it sounds like the lower frequencies have been boosted and the upper frequencies have been turned down.

 

So I thought I would have a listen to the previous version 0.7e (0.7.4) along with the new version 0.8 GCC and compare them to the original CD from which I ripped the music. The ripped WAV file played through version 0.7.4 sounded very close to identical to the original CD. The CD was played in a Sony SACD/CD transport via S/PDIF into the Weiss DAC2, and the ripped WAV file was played through firwire from the Mac into the DAC2 so it was easy to compare them.

 

Version 0.8 GCC on the other hand had exagerated bass most noticeable with the kick drum and bass chords, while the higher frequencies were overpowered by this bass, which interferred with the projection of voices and percussion, which lost their imaging as a consequence, becoming less defined.

 

This increased bass may be better in a system which is lacking bass response, but with full range speakers it is easy to discern this exagerated bass. I don't use tone controls or EQ, as I want the sound of the system to be close to the original source material.

 

Damien, I hope this feedback helps the most realistic sounding software player I've yet found get better & better. I don't know if you can get it to sound any better as it is close to the source material now, changes would need to be very subtle as it is nearly their.

 

[iamoneagain]

If your GCC version sounds bassy, how bad does the LLVM version sound? I haven't gone back to 0.7.4 but from memory, the GCC .8 version sounds the same.

 

Your description of your current GCC sounds exactly the way I feel about the LLVM version.

 

Edit: Just listen to 0.7.4 and it sounds exactly the same as GCC 0.8. Are you sure you're not somehow listening to LLVM version?

 

[Terry]

I would not call the latest GCC base heavy and I have a full range system. I would quickly say the latest fidelia with hog mode sounds awfully good. I think the .08 GCC sounds fine and the only thing I like to see is more ease of use and more body to the voices a less ethereal quality to the vocals. I am confident that by the end of beta it will be pretty special. I am liking some things about Decibel compared to it but it is close and of course since I have a license for Pure Music I am awaiting the release of the anticipated integer mode when it supposed to take a large jump in sound quality. I like all of those and the above mentioned fidelia which is for now my player of choice for now.

 

[wgscott]

I used 64-bit only binaries I compiled from svn with all 3 compiler options, and then the two versions of 0.8 on the download site. I simply cannot distinguish between any of them. Of course my system is limited, my hearing is bad, my dog is old, etc. But it does make me wonder if this is mass hysteria...

 

[Terry]

I think there are minor differences but I honestly could not tell you what is better and I might favor the a version/ I do think it is being greatly over blown as far as music delivery.

 

[iamoneagain]

Finally gave Fidelia another listen and it does sound pretty good but think GCC has just a little bit more sparkle and it's more tube-like.

 

Obviously Fidelia a better looking player but I hate that you can't drop and drag files into the player. You have to search for your artist each time or have playlists setup. So for now I'm sticking with Audirvana 0.8 GCC.

 

[Jud]

wgscott writes:

 

Of course my system is limited, my hearing is bad, my dog is old, etc.

 

I vote for the dog! ;-)

 

- though as I've said in another thread, the Peachtree isn't a favorite of mine. OTOH, my DAC is about 20 years old, so what do I have to brag about?

 

But it does make me wonder if this is mass hysteria...

 

There is the old Richard Feynman quote that the first rule is you must not fool yourself, and you are the easiest person to fool. But I do try to be careful when listening to test for SQ (otherwise I can just let go and enjoy the music). I did several listening tests between the gcc and llvm versions, and the results were always the same. You'll also notice upthread that whether the gcc version (several folks, maybe most?) is preferred or the llvm version (me), the descriptions of how the two versions differ are pretty consistent. So I'm thinking there's a decent chance that in this case there actually is a difference. (At least in some systems - maybe you have one of those systems Miska was describing in the other thread where everything sounds the same because it is perfect. ;-)

 

[wgscott]

If you have the identical code-base, and one compiler produces a binary that sounds different from a binary produced by another compiler, then at least one of the outputs cannot be bit-perfect, right? This is like the ultimate internal control. So either one of the compilers is producing a binary that yields some sort of inaccurate computation, or else there is no genuine difference. In other words, if the same code produces two different sounds on the same equipment, it must be because at least one compiler has introduced errors that no longer permit bit-perfect rendition.

 

[wgscott]

I've just put on the Tuvan Throat Singers in his honor...

 

[4est]

I have been trying to fool myself for years to no avail...

 

I do not understand squit about compiling software and it's net effect on how the resulting program runs. Seeing as how we don't know why/if players alter the sonics, are we in a position to say something cannot do it?

 

Bela Fleck and the Flecktones "Live at the Quick" featuring the throat singer Kongar-ool Ondar coming up!

 

[Jeff In San Diego]

...tin-foil-hat-guy on this one.

 

Knowing the basics of what a C compiler does, it is a real stretch to imagine some way in which the same C code compiled into two different executables for the same target machine (a Macintosh in this case) would give two apps that are different in a meaningful way.

 

The two apps created here would not necessarily be bit-identical. Two different compilers would have two different sets of op-code translation and two different sets of op-code optimization algorithms. They would likely create two executable files that differ in small ways.

 

They would, however, share the same logic, flow, and numerical-representation instructions. vis:

- get this number from "here"

- store it in this way with this many bits (floating point, 32 bit integer, etc)

- do this to it

- send it "there"

 

On one type of machine, any two executables that aspire to follow these steps with the same "here" (input file) will wind up with a bit-identical "something" in the location "there". The two apps could arrive at this point in very slightly different ways (two fundamental Intel op-codes swapped in order in a case where this makes no difference, for instance).

 

The two apps would not be bit-identical...BUT THEIR OUTPUT WOULD BE BIT IDENTICAL.

 

The ONLY way (and it is indeed a big stretch) for the resulting sound to be different, would be for these two executables, with their VERY similar but not identical coding structure, to have a different impact on the processor and it's ability to do the other parts of the sound rendering as they should be done.

 

In other words, a vanishingly small difference in TIMING of the execution of the Audirvana code and other code running on the system. We're talking ocean waves from the flapping of (high altitude) butterfly wings here people.

 

At some point in the development of Audirvana, I would invite Damien to make the following experiment (and make us all very red-faced). He should choose a week to take a much deserved vacation. At the end of this celebrated week, he should increment the version number of the app, and spin a new version and post it. No other changes; just change the version number and nothing else. The announcement post here should say simply "found and fixed a sound quality bug". Nothing more.

 

It would be interesting to see the comments on SQ changes for such an experiment.

 

My deepest apologies to the good people who think they hear a difference. As a card-carying member of the flat world club, I refuse to even compare them myself.

 

 

 

[wgscott]

... if one or the other compiler (or both) are making mistakes.

 

[CG]

I just know that this will come back to haunt me, but...

 

This is a mixed signal system. One of the implications of mixed signal processing is that the analog segment isn't 100% isolated from the digital part. (One wonders how that much isolation could even be possible outside of some imaginary computer model, but that's a different conversation.)

 

Anyway, if any combination of software application(s) on one computer running one version of the operating system causes the collateral effect of different analog noise signal spectra into the AC mains, into the DAC, and/or into the other analog components through common mode, normal mode, or radiative means, then you'll get a different analog output signal. That's even if the bit patterns are perfect, identical, and blessed by the Grand Wazoo.

 

We listen to analog signals.

 

I can and have measured this with real test equipment. (Disclaimer: This was largely done on much higher data rate and more complex signals than audio playback...)

 

Sorry, that's the way it is in the real world I inhabit. YMMV, etc.

 

 

 

[4est]

Good to see you again CG!

 

[Blu]

In my quest for the most realistic sound from my system, using a music server as source, I have tested different programs and methodologies and have found the following procedures to give the best results in transferring music from CD, so far.

 

Realistic in the sense that the players and their instruments and voices sound as if they are on a stage in front, to the sides and behind the speakers. Being able to not only hear the sound from guitar strings, for example, but also the fingers on those strings, and the sound of the body of the guitar. Also that metal guitar strings sound like the shimmer & twang of steel and not sound like nylon as I've heard on some systems. Then this image depends on a good recording, hence my desire for a faithful copy of the source.

 

RIPPING: I clean the CD playing surface with "Optimax", I then rip the CD in my MAC mini to WAV format, which I have found to give the most faithful sound to the original CD. I know there are problems with album art and metadata but remember the goal for me is the most realistic sound, if you include the album art as a .png or .jpg it will display in Audirvana. In some players you need to name the album art 'cover.jpg' or 'cover.png' for the program to pick it up. Also I store the albums in a folder under the name of the artist, for example MusicLeonard CohenThe Essential Leonard Cohen, Disc 1Names of tracks and cover.png file. This fits in with my choice of playing whole albums, if I'm having a party I will make up a playlist.

 

RIPPING SOFTWARE: I have found that 'MAX' gives the best sound, I'm using version 0.9.2 (January 2011) which works best with Snow Leopard on the Mac, I have also tried XLD. I have sent a donation to Stephen Booth who made 'Max' and also made 'Decibel', for his trouble.

 

I have experimented with the settings in 'Max' and have found the best results with the 'cdparanoia' ripper into WAV (Microsoft) format from 'libsndfile' as source, including the Paranoia setting, Error correction enabled, Full paranoia and skipping after 20 retries. I have also found that using 'one encoder' sounds more faithful to the original than using two. I know eveyone is in a rush today so they will use whatever is fastest, unfortunately gain in speed can have loss in quality.

 

In the MAX iTunes setting I have a copy sent to the iTunes library, as I also have an iPod Touch and a license for 'Pure Music'. The iTunes library is on an attached USB drive, which I originally used for music playback. I have recently bought a Network Attached Storage ('Seagate NAS 110') disc and now copy music from the USB drive to the NAS. The NAS is part of my wired network, and I have found the music sounds better coming from it than the attached USB drive. I had read in a thread on CA this is because this excludes any jitter transfer to the MAC.

 

PLAYBACK SOFTWARE: I have found 'AUDIRVANA 0.7e (0.7.4)' to give the most realistically faithful sound to the original CD so far, I have also tried 'Pure Music', 'Play', 'Decibel' and 'Fidelia'. I'm not saying these players are bad, I have found that their sound is not quite as realistic, it depends what you appreciate in music, and how picky you are, maybe I'm too picky.

 

I have experimented with the settings in 'Audirvana', and found the best results under Audio System, Exclusive access mode, Max I/O buffer size, Audio Filters; SRC libSampleRate converter, Best Quality setting and No upsampling, Daniel Weiss doesn't think Integer Mode is needed in the DAC2. The signal leaves the Mac mini over a firewire cable to the DAC.

 

A common thread going through this procedure is that all the software I'm using has Open Source components, including 'cdparanoia' for ripping, 'libsndfile' for the WAV music format, 'SRC libSampleRate' for conversion in the playback software which is compiled with the 'GCC compiler' (as 'wgscott' our favourite 'flat earther', noted is from the great Richard Stallman).

 

I didn't set out to use this software, it just happens that through experimenting, listening and reading this site, I have achieved my goal of most faithful reproduction of the source material (CD in this case), using software that seems to work together to give a great result.

 

I have gladly donated to Damien for his great Audirvana, and try to use links on CA where I can to help Chris Connaker, the creator of this great resource.

 

Enjoy the Music, that's what its all about,

Blu.

 

[Blu]

I wrote the comment "Maybe the answer is in the details", as I have commented in this thread previously, on my finding that particular versions of Audirvana sounded to me better than other versions and even other players. So I wrote this new comment to provide a consistent reason as to why this might be possible.

 

It is not only the player, but the whole procedure I use from the ripping the CD, to the music format, to the player that all contribute to what I'm hearing, and that if you don't follow this procedure you wont be hearing what I'm hearing. Make sense?

 

Enjoy the music,

Blu

 

[57gold]

Wondering why you haven't added RAM and SSD to your Mini?

 

Guessing with quality of rest of your system, you might hear an improvement. Gordon Rankin, who helped me get into computer audio and make some choices of gear and approach said "I wouldn't be sorry" for goosing up my Mini and he was right.

 

Upgraded mine at OWC with 120 GB SSD (@$225) and 8 GB RAM (@$110); since I am not proficcient in eletronics had OWC do it for $99 including roundtrip overnight shipping and it took 4 days.

 

Seems like a logical step in your "success is in the details" approach.

 

[Terry]

I think that doing that to you mini was a good thing. I like this player as I have said many times but I wonder if we are making a bigger deal over the compiler than we should be. I am sure in time it will get better and he was nice enough to let us have a few versions to test. I say remember this is beta and let it develop in its own way.

 

[iamoneagain]

I can clearly hear a difference between the two compilers and could live with the newer version but it's just a little off for me. I would probably end back with Amarra if I didn't have the GCC version to still use.

 

I finally went back and listened to Amarra again, which was my favorite sounding player before, and was surprised how close it and GCC version sounds. They are the only 2 players that bring out the highend sparkle while also keeping the sound smooth and tubelike. Of course I encountered an error with Amarra right off the bat, which is why I started searching other players in the first place.

 

[damien78]

Release 0.8.1:

• Special settings (in user prefs) for SPDIF bridge devices:
  
  Max sample rate limit
  
  • Additional latency when switching sample rate to allow connected DAC to synchronize

  [*]Bug Fixes:

  Stop (reset device) when exiting Audirvana while playing
  
  • Repeat mode and other playlist bug fixes
    

 

I've posted to audirvana.googlecode.com both llvm 2.0 and gcc builts for your continued A/B testings.

I'm still wondering which one will be the one for release 1.0

 

Regarding explanation of these impacts, I like CG's one: the different opcodes running in CPU generate different loads on the computer PSU that goes into mains. Then the DAC is fed by the same mains...

In addition to mains noise, there is also RF noise that is emitted by the wires going in/out of the computer.

So bit perfect is just the beginning of the story. Otherwise any transport would sound the same...

 

The funny thing is that Apple claims that llvm generates more optimized binaries, thus faster code that'll drain less on the mains. In a nutshell, Mr Apple says it should go in the direction of better SQ...

 

Damien

 

[Joebah]

So, for my set-up [V-Link (24/96) to V-Dac (24/192)] what settings should I use for the new bridge options in 0.8.1?

 

Thanks!

 

[Joebah]

Using the newest trial version... When I quit, there was a loud buzz coming through my speakers. Will try to restart and try again (download going on now).

 

[Blu]

You are right '57gold', upgrading to a Solid State Drive and taking RAM to the max is the next step. I have been thinking of doing these for a long time, since reading early on at CA here, of the positive benefits from Chris and Gordon of that approach.

 

Enjoy the music,

Blu

 

[Blu]

Thank you Jud for taking the time to provide the detailed info on using the compiler of your choice.

 

Now to sort out when to start downloading Xcode, as over 4GB is going to take a long time. Hopeing my modem doesn't drop out part way through.

 

Many thanks, and enjoy the music,

Blu

 

[57gold]

Made a contribution...2X upsampling to 88.2 sounds freakin' great.

 

A little Charlie Hunter Band with conga, vibes, thumping bass...total kick ass!

 

A little early on my praise for newest version: Selected a Brad Mehldau file and got clicking sound (like phono scratches); rebooted and selected another file, Steve Earle...more crackle/pop? Then took off upsampling...cracks even more.

 

Not sure I selected right preferences. Any ideas?