FLAC’s dirty little secret

First — and briefly, don’t worry ! — some personal historical background.

My journey in “computer audio” started many years ago with Logitech’s Squeezeboxes. From the very beginning, I looked for the best and more versatile file format. In those times, FLAC was not an option on Mac so I settled for AIFF, which was much better, musically, than ALAC, while preserving metadata, which WAV did not. When the time came to convert my library for use with “JRiver”, I was listening with a Bel Canto 3.7 and NuForce Ref 9 SE amps. I converted all the AIFF into FLAC and settled for the default compression level of “6”, as it was labelled “recommended”. My listening tests in these times didn’t indeed reveal any noticeable difference with AIFF and I rejoiced at the 50 % gain in file sizes. And forgot about it.

Fast forward to the autumn of 2021. After about two years with a Bartók (+ Rossini Clock), I could afford a Rossini. As I had the cash, I opted for the player over the DAC alone, considering the drive as a convenience. How surprised and puzzled I was, on the first listen, to discover that spinning CDs provided better results, more musical enjoyment than digital files — yes, even with Mosaic and MinimServer, I do play by the rules.

I couldn’t find any convincing, reasonable explanation… until I remembered this compression setting for the FLAC conversion and decided to do some tests. In short the results are… [Drum roll]… with compression level 0 (zero, the less compression), the sound is noticeably improved over compression level 6. I would say it’s probably equivalent to the CD and I’m very happy about it. My wife and I, we did some listening and comparisons with AIFF and WAV files ; we came to the conclusion that FLAC 0 was undistinguishable from AIFF and WAV and all these options are noticeably better than FLAC 6. I wouldn’t swear that it’s completely equivalent to the CD in all situations, as it would require very extensive testing with different recordings, different genres and so on, as you all know. What I can say, after days of listening to my converted music library, is that the music is more lively , more “life like”, got more of what we, audiophiles and music lovers, are looking for in our hobby. Anyway, I don’t feel the urge any more to reach for the CD — when I have it — as it was the case before conversion.

Let’s say I’m very happy with the improvement in sound quality that I could implement by converting the whole library. It’s also very satisfactory that it costs absolutely nothing moneywise ! To my ears, it’s a sonic improvement similar to upgrading cables, and you know how much this can cost !

So, in the end, I would advise anyone listening to local FLAC files to convert them into compression level 0. It costs nothing but the effort to do it — not much if you have the tools and the know-how — and either you have no gain or you have a very enjoyable gain musicwise. There’s nothing to lose, but a few megabytes of disk storage, depending on the size or your library — the difference in size between compression level 6 and level 0 is about 10 %. Seen the other way around : it’s not worth gaining 10 % disk space, nowadays, if it’s at the expense of the sound quality, especially if you’ve got gears in the dCS class !

Olivier :-{)

PS : the other components of my set-up :

  • Jeff Rowland 625 C2
  • Duevel Bella Luna
  • Isol-8 Substation Integra
  • all wired with Cardas Clear but the power cords into Integra and into Jeff Rowland (Clear Beyond)

Nice info. As a Microsoft Windows user, I’ve been ripping my CDs to WAV since Exact Audio Copy first came out.

At that time, i had trouble embedding metadata tags into the files, so I began ripping to FLAC, with compression level zero.

While i have not tried other compression levels, to my ears, FLAC zero sounds indistinguishable from WAV.

Thanks Olivier and Leonard!
I “used to know” that wav sounded better than flac but forgot about it when I noticed how good my Network Bridge played flac files.
After reading your posts I made some experiments with my present rig and now I have put JRiver to work on my flac 16-files - some 30K on my Synology NAS :sweat_smile:

Respectfully guys, I have to disagree with premise of this thread.

If you do an objective test, you’ll realise that there’s actually no difference. If there was indeed was difference, it would suggest that dCS folks have done a poor job with such a simple task of decoding FLAC without impacting sound quality. Do you really believe that? :laughing:

In any case, this isn’t just rhetoric, I’ve done the test before, and there’s even a thread about this just over a year ago, I recommend reading it in full (or from post #47 linked here);

The dCS architecture has FLAC decoding occurring on a separate compute board (S800) that’s isolated from the rest of the DAC. The integrated nature of the Rossini and Bartok makes independently analysing this Compute board tricky, but with the Vivaldi Upsampler and the dCS Network Bridge, it’s very easy prove objectively, beyond any doubt, that theres’ absolutely no difference to the decoded bitstreams regardless of the FLAC compression levels, and consequently, no difference to the sound quality.


Thank you for your input, Anupc! It is always interesting to take part of opinions based on theory – it can help you look for answers in the right places. However, in the final (temporary) analysis, I always trust my empirical observations more. If the theory is not supported, I have to look for explanations elsewhere. And these explanations should build on observations made rather than thoughts about what things ideally should be.

In the case of comparing flac-files with compression level 0 to level 6, I made a simple experiment before taking action. First I chose five pieces of music that have differences in character, style and performance. Then I listened to and compared the flac0 to the flac6 of each of these pieces of music. I did this in my desktop rig, my headphone rig and my main rig. The last two situations makes use of a dCS Network Bridge into two different dacs. In all three situations my listening experience made it clear to me that there was a difference in sound quality between the level 0 and level 6 of compression. At all times I preferred listening to flac0 files.

So for me, the choice of action was simple: start converting all my flac6 files to flac0. And that is happening right now.

Im trying to understand. I think what youre saying is you are taking a file that is already compressed to level 6… taking that compressed file and by rerecording that file in compression 0 format yielding a better file? Am i misreading this?

He’s changing the compression level in the encoding phase. Since FLAC is a lossless format you can re-encode as many times as you like with no degradation to the original data.

The decoded content of the level 0 and level 6 files is exactly the same. The theoretical difference is how much processing work the CPU has to do when decoding the file. Genearlly this would be higher for higher levels of compession.

That being said, I’m with Anup on this one, I don’t believe it could possibly make a difference in the sound.


The problem with FLAC and ALAC is that they are compressed, although they are both technically bit-perfect but compression can cause delays because your device has to decompress the PCM data in real time, so there will always be a latency. The higher the compression level, the greater the latency. This is also why ALAC always sounds worse than FLAC because its compression algorithm is designed to maximise the compression ratio.

That is broadly so but the delays ( or their outcome) is not always apparent. I have some comparative test files without compression and at various compression levels. I could easily hear some degradation using earlier dCS ranges (in my case, Elgar and Paganini). However I cannot hear any with Vivaldi. So perhaps later implementation is better - note that Vivaldi v.2 runs at twice the speed of earlier iterations of Ring DAC.

When ripping I choose FLAC 0 which minimises compression but maintains the embedding of metadata.