KaktoR, send me a sample of the files via private message or attach them here. That will make things easier, and I'll be able to work on the ADPCM model – I can see there's still some room for it!
EDIT
And perhaps your samples are not being detected as ADPCM (VAG/PS-ADPCM, DSP, Xbox ADPCM, or a sound bank).
My samples (IMA and MS ADPCM in RIFF WAV, files of one kind put together; Ryzen 7 5800X3D, 16 threads, every output decompressed and compared):
Code:
IMA ADPCM WAV (11 767 582 B) | MS ADPCM WAV (11 827 746 B)
compressor size ratio comp decomp | size ratio comp decomp
bcm 2.03 (default) 7 999 145 67.98% 0.84 s 0.83 s | 6 485 638 54.83% 0.78 s 0.72 s
bzip3 -j 16 8 001 713 68.00% 1.15 s 1.13 s | 6 489 847 54.87% 1.05 s 0.97 s
rofz 0.17.2 -l4 7 779 664 66.11% 6.46 s 0.21 s | 6 556 774 55.44% 8.10 s 0.19 s
rofz 0.17.2 -l8 7 061 197 60.01% 7.29 s 0.28 s | 6 048 704 51.14% 9.86 s 0.27 s
rofz 0.17.2 -l8 -na 7 589 486 64.49% - - | 6 330 231 53.52% - -
On IMA / MS ADPCM WAV,
-l8 is 11.7% / 6.7% smaller than BCM, and even without the audio model (
-na, plain LZ) still 5.1% / 2.4% smaller. On your file BWT wins over
-l8 by 4.4%, so it is most likely an ADPCM format rofz does not detect yet.
Or perhaps it’s a bug from 0.17.0 related to the WAV / ADPCM header.