On 17 September 2012 17:16, Paul Giblock <[email protected]> wrote: > > On Sep 17, 2012 8:04 PM, "Henner Zeller" <[email protected]> wrote: >> >> Hi, >> On 16 September 2012 09:10, Fons Adriaensen <[email protected]> wrote: >>> >>> On Sun, Sep 16, 2012 at 12:46:42AM -0700, Henner Zeller wrote: >>> >>> > Over the week or so, I've build some fuse filesystem filters flac files >>> > on-the-fly >>> > using the zita convolver. This makes it real easy to play >>> > around with files and filters. In particular with systems that >>> > otherwise >>> > don't support stream convolving, but just can read files. >>> >>> Interesting project... >>> >>> One thing: when skipping forward (or seeking in general) >>> there's no need to convolve all data up to that point. >>> If P is the new position and L is the lenght of the IR, >>> then just call Convproc->reset(), seek the input file to >>> P - L (or 0 if P < L), and convolve up to P. You're >>> computing a FIR filter, so no more history is required. >> >> >> I know. The only reason for doing what I am doing is entirely because of >> the filesystem semantics. >> >> On the filesystem level, I don't 'see' seeks to audio frame positions but >> _only_ file positions. Since the input and the output is a compressed flac >> file, the relationship cannot be predicted - this is why I have to walk up >> to that point to present a consistent view of the file (in particular if >> that file is mmap()ed or if users jump around in it, this is important). >> (this would be different if input and output was wav where the >> relationship is clear; but wav doesn't provide such a rich header format, so >> that the user in media server that display tags and images would be bad). >> >>> >>> If the seek involves an output discontinuity for the user, >>> you could even just call Conproc->reset(), seek to P, and >>> proceed normally. The result will be as if there was only >>> silence before P. >>> >>> Ciao, >>> >>> -- >>> FA >>> >>> A world of exhaustive, reliable metadata would be an utopia. >>> It's also a pipe-dream, founded on self-delusion, nerd hubris >>> and hysterically inflated market opportunities. (Cory Doctorow) >>> >> >> >> _______________________________________________ >> Linux-audio-dev mailing list >> [email protected] >> http://lists.linuxaudio.org/listinfo/linux-audio-dev >> > > Have you considered OggPCM? Perhaps it has a constant enough bitrate while > still providing rich tags. Not sure how well supported it is though...
Thanks for the tip. I think OggPCM sounds like a good way to represent the data with predictable seek behavior. Right now, i didn't run into the problem for flac files with my current usage pattern (=not much seeking :) ) though, but it might be a good feature in the future. The beauty if encoding flac -> flac right now is, that I can just provide the original headers verbatim to the new file, so I don't have to worry about extracting the header and re-coding it into a different header format and probably loosing half of the interesting information on the way (e.g. libsndfile does not support cover-pictures). (All my audio files are flac anyway). Where it immediately can be useful: when reading ogg-vorbis files. I do re-coding ogg-vorbis -> flac right now in Folve, because apparently libsndfile needs to read the full input before it can generate the compressed output -- which is not what I want, so I encode as flac to have the immediate 'streaming' capabilities. In that case using the same ogg container but having the data in PCM might be the better choice; Have to check that libsndfile supports the OggPCM combination. -h _______________________________________________ Linux-audio-dev mailing list [email protected] http://lists.linuxaudio.org/listinfo/linux-audio-dev
