Wav File
in [DSP]
|
Prev: IIR and FIR filters
Next: Possible repost - Using least squares cubic spline fitting asa "filter", what is analogus to Nyquist criterion?
From: Richard Dobson on 15 Nov 2009 13:36 Jerry Avins wrote: ... >> And I will quietly note that ~nobody~ else responding to the OP >> offered this obvious (to those who know...) and simple piece of >> information! > > Because until the OP understands what a .wav file is, the simple (and > excellent) piece of information won't do him any good. He first needs to > know that PCM _is_ integer. Then, if he really needs floating point, he > needs to learn his programming language well enough to change types. > Or use (or copy to) a floats WAVE file - then you can just read and process. Still PCM (at least, in the sense of "not compressed"), or so I'm told. Lots of integer types in WAVE and AIFF files BTW: covering all byte-lengths from 1 to 4 inclusive; 2 and 3-byte words of course the most common by far. And with WAVE_EX, you can in principle have 20bit words inside a 24bit container. So anyone parsing a PCM WAVE file still needs to check if the data is integer or floats, what size of integer/float the file contains, etc etc etc. It is a ~lot~ less work, and a whole lot safer, just to use an established library to do the job. Richard Dobson
From: Jerry Avins on 15 Nov 2009 14:39 Richard Dobson wrote: > It is a ~lot~ less work, and a whole lot safer, just to use an > established library to do the job. Yes. I stand corrected and abashed, but the OP still needs to know what a .wav file is -- in particular, _his_ .wav file. He can use libsndfile to read the header. (Do you think he knows about headers?) Jerry -- Engineering is the art of making what you want from things you can get. �����������������������������������������������������������������������
From: Richard Dobson on 15 Nov 2009 16:24 Jerry Avins wrote: > Richard Dobson wrote: > >> It is a ~lot~ less work, and a whole lot safer, just to use an >> established library to do the job. > > Yes. I stand corrected and abashed, but the OP still needs to know what > a .wav file is -- in particular, _his_ .wav file. He can use libsndfile > to read the header. (Do you think he knows about headers?) > > Jerry The only essential point to make in answering that question is that a WAVE (or AIFF) file does not have a fixed-length header. The OP shoudl disregard any suggestion to the contrary. You can't take a short cut and say 'the audio starts at byte 34' or whatever. Well, you could, just, in writing a file, but not in reading one. There are unfortunately many examples (including many self-styled "documentation" pages) on the net where that assumption is made, implicitly or explicitly (the original matlab implementation of "wavread" suffered just this problem, among other things). I long ago lost count of the number of unix programs especially that broke simply because I gave them a WAVE file with a longer header than the theoretical minimum. All other things being equal, there are only two primary types of programmer who should really commit the time and effort to delve into the minutiae of such file formats - writers of soundfile handling libraries such as Eric with libsndfile (which as I am sure he will confirm has taken years of work to reach its present extremely efficient, accurate and comprehensive state), and people who design file formats (and thus need a custom soundfile library they can freely "hack"); which is what I have done from time to time (e.g. for phase vocoder analysis files, and for Ambisonic B-Format). Richard Dobson
From: Erik de Castro Lopo on 15 Nov 2009 16:36 Jerry Avins wrote: > Richard Dobson wrote: > > > It is a ~lot~ less work, and a whole lot safer, just to use an > > established library to do the job. > > Yes. I stand corrected and abashed, but the OP still needs to know what > a .wav file is -- in particular, _his_ .wav file. He can use libsndfile > to read the header. (Do you think he knows about headers?) Quite honestly, manually read WAV file headers is something to be avoided unless you have a very good reason not to. WAV files have a rather poor sepcification and there have been many, many programs over the years that produced odd or non-standard files. libsndfile's WAV parser is full of little workarounds and special cases for some of these non-standard files. However, for those interested, the sndfile-info program which comes with libsndfile dumps all the header information in a human readable form liek this: erik > sndfile-info drumset01-sep08wallywoods-zettberlin.wav Version : libsndfile-1.0.21pre1 ======================================== File : /home/erik/drumset01-sep08wallywoods-zettberlin.wav Length : 270608864 RIFF : 270608856 WAVE fmt : 16 Format : 0x3 => WAVE_FORMAT_IEEE_FLOAT Channels : 2 Sample Rate : 96000 Block Align : 8 Bit Width : 32 Bytes/sec : 768000 fact : 4 frames : 33826097 PEAK : 24 version : 1 time stamp : 1221772493 Ch Position Value 0 12307085 0.817219 1 12307085 0.712256 data : 270608776 End ---------------------------------------- Sample Rate : 96000 Frames : 33826097 Channels : 2 Format : 0x00010006 Sections : 1 Seekable : TRUE Duration : 00:05:52.355 Signal Max : 0.817219 (-1.75 dB) Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/
From: Erik de Castro Lopo on 15 Nov 2009 16:45
Richard Dobson wrote: > All other things being equal, there are only two primary types of > programmer who should really commit the time and effort to delve into > the minutiae of such file formats - writers of soundfile handling > libraries such as Eric with libsndfile (which as I am sure he will > confirm has taken years of work to reach its present extremely > efficient, accurate and comprehensive state), and people who design file > formats (and thus need a custom soundfile library they can freely > "hack"); which is what I have done from time to time (e.g. for phase > vocoder analysis files, and for Ambisonic B-Format). I agree with this completely. Reading most files is relatively trivial but there are a huge number of tiny variations, some within spec and some no that make reading as many files as possible a pain in the neck. On a side note, libsndfile has been publicly available for a little over 10 years. The 10 year retrospective is here: http://www.mega-nerd.com/erikd/Blog/CodeHacking/libsndfile/ten_years.html libsndfile actually grew out of code I originally wrote in 1992, when i decided that microsoft's API for reading WAV files was simply way too horrible to use. Erik -- ---------------------------------------------------------------------- Erik de Castro Lopo http://www.mega-nerd.com/ |