The package can be downloaded here.
lpc10-1.5.tar.gz (800 KB gzip'ed tar file)
lpc10-15.zip (940 KB ZIP file)
To unpack the gzip'ed tar file on a Unix machine, you can do the following two steps:
gunzip lpc10-1.5.tar.gz tar -xf lpc10-1.5.tarI have used free implementations of the programs gzip and tar on a Macintosh platform as recently as 1995. Among other places, these can be obtained from the Stanford Info-Mac archive.
If you want to be able to use the LPC-10 compression for any of TI's speech synthesis chips, or you'd like a nice GUI running under MS Windows, I'd recommend looking at Quadravox's web pages. If they don't have what you want listed on their web pages, send them E-mail and see if they are willing to accomodate you. I have no affiliation with this company at all, and I'm not even a happy satisfied customer of theirs. I have corresponded with one person who found their products useful, in a place where this LPC-10 software was not as useful.
When I modified this software, my intent was not to alter the speech processing algorithms of the original code at all, and I believe I have achieved that goal. I only modified the way that the encoding and decoding functions were called, checked carefully over the code for potential buffer overrun problems and similar bugs (I only found one spot that increasing a buffer size was possibly useful) by hand and using Purify, and removed tracing code. I also made the decoding routines return the same number of samples on every call, instead of a variable number, but the only effect that has is to increase the delay of the decoding routine by at most one frame time (22.5 ms). The actual samples returned should be the same as the original LPC-10 code.
So, I have some knowledge about the details of the code, but I wasn't attempting to learn all of the speech processing algorithms as I was doing it. If you want to know how LPC analysis works, I'd suggest consulting a speech processing book. If you want to know more about LPC-10's algorithms in particular, there are several research articles mentioned in the file FAQ in the distribution (the first half of this file is about the CELP 4800 bps standard -- the second half has references about LPC-10). If you do not know where to get copies of these articles, just go to the nearest library and show the librarian the journal article citations, and they should be able to help you.
Another potential place to learn more about the internal workings of the LPC-10 speech coder is a MATLAB implementation of the LPC-10 coder and some accompanying training offered on the web page of Andreas Spanias. I haven't tried out this code myself, but it might be just what you're looking for.
The following book does not have details on the LPC-10 speech coder itself, but it does have a wealth of information on many techniques used in many speech coders. I heartily recommend it, as well as the second book.