Convert music to FLAC format using CUDA

    I recently came across an encoder working with the FLAC format and transferring part of the encoding process to the GPU using CUDA. The program is called FlaCuda and you can get it here . Despite being written using .NET, it works great with Mono. In any case, I did not have any difficulties with its use on Ubuntu 9.10 amd64.

    The program works pretty fast, while compressing the same audio file flac, flake and flacuda showed the following results (amd x2 4200+, geforce 9800 GT):

    programcoding timefinal file size
    flac127 p.425.3 Mb
    flake106 p.425.9 Mb
    flacuda24 c.424.1 Mb


    The compression level for all programs was set to 8. We get a solid five-fold increase in speed. True, such a gain can be achieved only on long files. On 3-6 minute recordings, the encoding time is reduced by about half.

    The program can take data from standard input and output to standard output, which allows you to stick it into the pipeline. For example, I quite successfully modified cue2tracks to work with flacuda.

    Now about the cons. The program is built on the basis of the flake codec, so it can’t compress 24-bit sound. Very disappointing. I did not notice any more shortcomings.

    The author himself warns that the program is still in the alpha stage, so it is advisable to double-check the received files. For example, you can overtake the resulting .flac back into .wav and compare it with the original one by calculating the checksum of both. For development, you can follow the forum www.hydrogenaudio.org , and there you can contact the author.

    And finally - a bonus for linux users, a modified cue2tracks script, which adds support for FlaCuda. For the script to work, you need to place the executable file "CUETools.FlaCuda.exe" in "/ usr / bin" under the name flacuda, and the dll libraries in "/ usr / lib".

    Upd. For Gentoo users, powerman has added packages for flacuda and the modified cue2tracks to the powerman overlay .

    Download script.

    Also popular now: