Best compressed audio?

[smack]

Thanks for this link! I really like this QOA codec and its decidedly simple but effective design.


The author freely admits that it decodes slower than ADPCM.

As ross said, it's probably too much for the 68000, because it needs (at least) 4 muls per sample, see function

qoa_lms_predict

https://github.com/phoboslab/qoa/blob/master/qoa.h

Like for other time domain codecs, the heart of QOA is in the prediction of the next sample. QOA uses a “sign-sign” Least Mean Squares Filter with 4 weights (“taps”). The prediction is simply the sum of the 4 weights multiplied with the 4 previous samples.

[Karlos]

Originally Posted by ross

Yes, it may be adapted but it definitely wouldn't be as fast as an ADPCM decoder.
From a quick look the only sticking point would be a multiplication per sample (data structures designed for 64-bit architectures can simply be redesigned).

For sure for a (fast?) 020+ machine.

This is why I went for the simpler ADPCM style with my first attempt: the decompression was just based on addition. The attempt to improve the quality was just to find the best N delta values and, along with an uncompressed starting sample make that the frame. In hindsight, it's debatable how good it was but it seemed ok at the time. It allowed me to decompress and mix a lot of stuff just on the fly. Back then I was a bit obsessed over the size of my 040 caches so what I used to do with anything steaming like this was to have a cache aligned workspace that I could move16 data into and out of, so that the caches aren't flooded.

[No.3]

Originally Posted by Karlos

... The attempt to improve the quality was just to find the best N delta values ...

Interesting approach, but makes the compression complicated i.e. it will need a lot of time?!

Anyway, at least for my ears, the exponential delta compression from the Aminet wavepak archive sounds far much better than the fibonacci delta compression.

[Karlos]

Originally Posted by No.3

Interesting approach, but makes the compression complicated i.e. it will need a lot of time?!

Anyway, at least for my ears, the exponential delta compression from the Aminet wavepak archive sounds far much better than the fibonacci delta compression.

Compression time was ok at least on 040. Decompressing mattered a lot more to me at the time.

[smack]

Originally Posted by Karlos

Compression time was ok at least on 040. Decompressing mattered a lot more to me at the time.

Xtreme Quality IMA-ADPCM Encoder / Decoder - https://github.com/dbry/adpcm-xq

Another example of this "asymmetric" idea - compress once on a fast machine and then play even on old / slow machines.

[paraj]

Just for fun I combined ADPCM, a bit of extra compression and ideas from http://eab.abime.net/showthread.php?...57#post1604757 to cram a 5 minute track onto a disk (only at 7813Hz though). In the zone as cigar_altmfm.adf.

It's a bit rough in places, but for what it's doing I think it's OK. Sample rate can be traded for length, so at 3 minutes you should able to get ~13KHz.