the penguin

Losslessly Compressing DSD files with Wavpack

One of the biggest disadvantages of DSD files is their size. Most of the users who like DSD format, keep large SACD ISOs or split them to .dsf/.dff files. In both cases, the result is horrible space-wise as the majority of music albums need more than 1.5GB (only for the 2 channel files).

It was not long ago that Wavpack project came with a solution to the above: a lossless compression algorithm for dsf/dff files, resulting in up to 60% smaller files.

In order to test it, I quickly installed wavpack:

pacman -Sy wavpack

and chose a SACD ISO file for my test. The size of the ISO was 3.5GB. After the extraction of 2 channel dsf files (who needs 5.1?!?), the total size was 2.6GB.

The next step was the wavpack compression. Reading the man page of wavpack command, I noticed some interesting options:

       delete source file if successful (use with caution!)

       fast mode (fast, but some compromise in compression ratio)

       high quality (better compression ratio, but slower encode and decode than default mode)

       very high quality (best compression, but slowest and NOT recommended for use on portable
       playback devices)

       import applicable tag items from ID3v2.3 tag present in DSF files into APEv2 tag (if there
       are > 1 MB cover images present add --allow-huge-tags to include them, and -r if you do not
       want large images appearing twice in the WavPack file, although this will remove the entire
       ID3 tag wrapper)

The first test I did was very basic and with standard compression:

wavpack *.dsf

The resulting .wv DSD files had a total size of 1.6GB.

Now it was time to test the high quality compression option:

wavpack -hh *.dsf

To my surprise, the resulting .wv DSD files had a total size of 1.2GB!

It was time to test these files with MPD. I quickly rebuilt Archphile packages with wavpack support and tested the highly compressed files (as this is the most CPU hungry scenario) using the Archphile/Odroid C2 combo as a transport. In my configuration, MPD is using only one and dedicated Odroid C2 CPU core. The decoding of the .wv file needed approx 25% of this core.

The most important part of this test, was to check if the files were recognized as DSD files during playback:

cat /proc/asound/card*/pcm*p/sub*/hw_params

the output was:

format: DSD_U32_BE
subformat: STD
channels: 2
rate: 88200 (88200/1)
period_size: 11025
buffer_size: 44100

My transport worked as it should in Native DSD mode.

The last test was to see if I can have my .dsf files back:

wvunpack *.wv

I got the sha256sums of the extracted dsf files, and I compared with the sha256sums of the initial .dsf files and they were exactly the same. Success!

I am sure that wavpack compression for DSD files is here to stay. It’s:

In other words, It’s what .flac is for .wav files. What remains to be seen is if the community will adopt it and make it the DSD standard for computer audio.

#linux audio #dsd #mpd #wavpack