Comparison of compression ratios between FLAC --best and TAK -p4, 5,074 songs compared Options

[skamp]

This post made me wonder whether I had over-estimated TAK's compression capabilities, so I decided to run a benchmark on my entire music collection.

I used caudec to transcode my 5,074 FLACs (--best, version 1.2.1) to TAK (-p4, version 2.2.0), one album at a time. All songs are CD audio (16 bit / 44.1 kHz, stereo). I measured the size in bytes of all the files comprising the album, as WAV, FLAC and TAK files. I then calculated the compression ratio of the FLACs and the TAKs compared to the WAVs, as a percentage. Finally, I calculated the difference in ratios between FLAC and TAK: that value (also a percentage of the WAV file) is always negative, meaning TAK compressed better than FLAC in all cases. Here are the results, as a CSV file.

On average, TAK compressed 2.19% better than FLAC. Here's a list of the top 10 albums where the difference was the most significant:

• Original Pirate Material by The Streets: -4.9%
• Flat Beat (CD single) by Mr. Oizo: -4.6%
• Tubular Bells 2003 by Mike Oldfield: -4.6%
• Mothership Reconnection (CD single) by Scott Grooves: -4.5%
• François by Desireless: -4.4%
• The Man-Machine by Kraftwerk: -4.3%
• Sublime by Sublime: -4.1%
• Doggystyle by Snoop Dogg: -4.1%
• Raising Hell by Run-D.M.C.: -4.1%
• Wish You Were Here by Pink Floyd: -4.1%

Note: I used -p4 with TAK simply because I forgot about the existence about -p4m. I later tried it on a couple albums though, and while TAK -p4 is faster than FLAC --best, TAK -p4m is slower. Improvement in compression is unlikely to be much larger than 0.2%, from what I can tell.

This post has been edited by skamp: Jun 25 2012, 09:44

[Porcus]

I then calculated the compression ratio of the FLACs and the TAKs compared to the WAVs, as a percentage. Finally, I calculated the difference in ratios between FLAC and TAK: that value (also a percentage of the WAV file)

You've miscalculated a bit (see below). To correct this, calculate the ratio TAKbitrate/FLACbitrate, subtract 1 and then multiply by 100 %. You will probably get negative 3 point something.

Your figure measures an improvement of 2.19 percentage points, not 2.19 percents.
Percents: improvement over what-to-improve (i.e. the FLAC file).
Percentage points: difference between percents.
The most crucial point is that all comparisons use the same yardstick (and I suppose that most should, and most will, choose the percents).

This post has been edited by Porcus: Jun 25 2012, 10:49

[skamp]

TAKs are on average 96.48% the size of the FLACs, so that's a 3.52% improvement over FLAC. That figure looks better indeed.

Edit: I've never used a spreadsheet in my life, so if you have some Excel-fu, feel free to download the CSV and calculate correct individual values for the "compression improvement" column by using the bytesize columns.

This post has been edited by skamp: Jun 25 2012, 11:19

[lvqcl]

WAV bytes, sum: 227851632252
FLAC bytes, sum: 141786750976
TAK bytes, sum: 136796209152

(TAK/FLAC)-1 = -3.52%

{too slow again }

This post has been edited by lvqcl: Jun 25 2012, 11:18

[skamp]

Revised CSV with the last column showing the compression improvement of TAK over FLAC, in %. My BASH-fu is stronger than my spreadsheet-fu

This post has been edited by skamp: Jun 25 2012, 11:32