Show HN: I told AI to write FLAC/MP3/DST decoders, got ~2x faster than FFmpeg
Show HN: I told AI to write FLAC/MP3/DST decoders, got ~2x faster than FFmpeg
Georgios Kakavelakis released three dependency-free C++17 audio decoders (MP3, FLAC, DST) written by Claude under his direction, used in his Orpheic music player and published under the MIT licence. Measured single-threaded against FFmpeg on an i9-14900K, Apple M4 and Raspberry Pi 5, FLAC ran 2.37x, 3.45x and 1.58x faster, the MP3 fast path 2.25x, 2.76x and 2.27x, and DST 2.01x, 2.06x and 2.89x (stereo), with FLAC and DST output bit-identical to FFmpeg's. The author notes MP3's exact path is several times slower than FFmpeg, DST was measured on excerpts that cannot be shared, and the code was checked against other decoders and by separate AI reviewers for copying.
Source: Hacker News AI · github.com