About
I've built a brute-force QOA/SEA encoder, which can be used to create high-quality QOA-encoded files, although very slowly.
The encoder implements the 1-, 2- and 3-bit versions of the QOA codec, although the QOA file format is supported rather poorly.
Since the encoder repeatedly brute-forces the best encoding within a given lookahead, the encoding can take multiple minutes/hours/... depending on the lookahead (scales exponentially).
The encoder is included in this repo: https://gitlab.com/BlobCodes/etdaw
The program architecture wasn't designed to be a QOA encoder, but a playground for creating entirely new codecs instead.
That's why the program may be a bit unintuitive to use right now or deliberately misses some QOA-specific optimizations.
Example (QOA 3.2 bits)
This screenshot shows spectrograms of the encoded outputs qoaconv and the brute-force encoder (using --quality 6 --metric RMS --format chunked-qoa) produce, showing that the brute-force encoder's output has noticably less noise in the higher frequencies.
The audio sample used was "Unknown Mortal Orchestra - Hunnybee".
Example: #69 (QOA 3.2 bits)
This encoder could be used to fix cases where qoaconv doesn't produce good/usable results.
For example, here is the audio sample from #69 encoded with the brute-force encoder (-m rms -f chunked-qoa -q 10):
About
I've built a brute-force QOA/SEA encoder, which can be used to create high-quality QOA-encoded files, although very slowly.
The encoder implements the 1-, 2- and 3-bit versions of the QOA codec, although the QOA file format is supported rather poorly.
Since the encoder repeatedly brute-forces the best encoding within a given lookahead, the encoding can take multiple minutes/hours/... depending on the lookahead (scales exponentially).
The encoder is included in this repo: https://gitlab.com/BlobCodes/etdaw
The program architecture wasn't designed to be a QOA encoder, but a playground for creating entirely new codecs instead.
That's why the program may be a bit unintuitive to use right now or deliberately misses some QOA-specific optimizations.
Example (QOA 3.2 bits)
This screenshot shows spectrograms of the encoded outputs qoaconv and the brute-force encoder (using
--quality 6 --metric RMS --format chunked-qoa) produce, showing that the brute-force encoder's output has noticably less noise in the higher frequencies.The audio sample used was "Unknown Mortal Orchestra - Hunnybee".
Example: #69 (QOA 3.2 bits)
This encoder could be used to fix cases where qoaconv doesn't produce good/usable results.
For example, here is the audio sample from #69 encoded with the brute-force encoder (
-m rms -f chunked-qoa -q 10):