keith preston a ?crit : > Ok, so I've been busy hacking away at getting the libspeex fixed point > resampler into pulse audio. Here is a patch that barely makes it work in > a very hacked way. You should be able to apply this to the tarball > 0.9.5sources. If you get it to build (I found a few problems with > types so you > might have to comment uncomment a couple of lines) you can run the > pulseaudio daemon with the parameters --resample=libspeex:0 where 0 > can be > 0-10 and represents quality. A few notes Cool! > It only does re sampling and not channel doubling (I have some code in > there, but it doesn't really work and probably is really bad) > > There is a annoying constant clicking, which I'm sure is just a bug I > overlooked. (Jean Marc?) Oops, I think you've hit a bug in the short->float converter (when using the _int() version but compiling as float) and which I fixed yesterday. Updating svn should do it. > I have this weird issue where the re sample will return an out length that > is not divisible by sample_size * channels. It tends to be odd. What > does this mean? It means that the out_len is the number of samples *per* channel and so is the in_len. i.e. if you have 4 channels of 1024 samples each, the len is still 1024 (not 4096). Jean-Marc
