VoxForge
VoxForge
I downloaded http://www.repository.voxforge1.org/downloads/Main/Tags/Releases/0_1_1-build726/Julius-3.5.2-Quickstart-Linux_(0_1_1-build726).tgz, and, per the README, ran
./julian -input rawfile -C julian.jconf
It asks for a filename. I put in a .wav file of an audio recording, but apparently not the kind of .wav it wants. It comes back with:
adin_file: channel num != 1 (2)
Error: failed to read q-themes-1a.wav as a wav file
*** glibc detected *** julian: corrupted double-linked list: 0x0a068350 ***
(and the backtrace and memory map of shared libraries).
Any way to debug? It crashes on all the .wav's I tried. When I use -input rawfile -filelist myfile.wav (same input file), it doesn't crash, but it spews out "Error in reading speech data" many, many, many times and nothing of use that I could find.
But reading the other posts on the forum, I wonder if it is really up for transcribing speech, anyway. Does that remain a chimera? The other packages I found (CMU Sphinx, ISIP) did not even seem to be this far along; as far as I could tell, they were toolkits with no realistic chance of transcribing an audio file.
Any clues to either working with julian/julius or other packages to try would be greatly appreciated.
Thanks,
--- (Edited on 6/22/2010 1:37 pm [GMT-0500] by karlberry) ---
> 1adin_file: channel num != 1(2)
It tells you are trying to decode stereo file while mono is required. You need to convert file to mono using sox for example
> Any way to debug?
Debug is usually done with debugger. You can find debugger manual on http://www.gnu.org/software/gdb
> then I use -input rawfile -filelist myfile.wav (same input file), it doesn't crash, but it spews out "Error in reading speech data
File list argument is supposed to be a text file with the list of files one per row, not the wav file itself.
> But reading the other posts on the forum, I wonder if it is really up for transcribing speech, anyway. Does that remain a chimera?
It depends on the amount of effort you want to put into it.
> The other packages I found (CMU Sphinx, ISIP) did not even seem to be this far along; as far as I could tell, they were toolkits with no realistic chance of transcribing an audio file
You probably need to put more time on investigation of the state of the things.
--- (Edited on 6/23/2010 04:13 [GMT+0400] by nsh) ---