LPC-10 vs. Texas Instruments LPC

Started by mhavoc October 20, 2006
Does anyone here possibly have detailed information on Texas
Instruments implementation of the LPC algorithm for speech compression?
I am in the process of writing an LPC-T1 and LPC-10 encoder/decoder
library in C# and I have Andy Fingerhut's LPC-10 code (which is in C,
converted from Fortran). From the data manuals I have, LPC-10 has 53
bits per frame plus a sync frame. TI LPC has 49 bits (no sync) plus a
'repeat' bit. The other bit differences are in the coefficient lookup
tables sizes it appears.

I have several of the TI TMS5XXX series datasheets on my website at...

http://www.gamearchive.com/General/Data_Sheets/

Before I go into severe specifics, I just wanted to know if anyone here
might be willing to help me with this topic or had more info.. 

thanks

mhavoc wrote:
> Does anyone here possibly have detailed information on Texas > Instruments implementation of the LPC algorithm for speech compression? > I am in the process of writing an LPC-T1 and LPC-10 encoder/decoder > library in C# and I have Andy Fingerhut's LPC-10 code (which is in C, > converted from Fortran). From the data manuals I have, LPC-10 has 53 > bits per frame plus a sync frame. TI LPC has 49 bits (no sync) plus a > 'repeat' bit. The other bit differences are in the coefficient lookup > tables sizes it appears. > > I have several of the TI TMS5XXX series datasheets on my website at... > > http://www.gamearchive.com/General/Data_Sheets/ > > Before I go into severe specifics, I just wanted to know if anyone here > might be willing to help me with this topic or had more info..
A quick comparison shows not much in common, except for the basic scheme of using LPC. The quantisation of every parameter is different. There seems to be no special unvoiced processing in the TI LPC. Steve
I just found Phil's HawkVoice library last night after searching the
comp.dsp archives and I think that even tho it doesn't appear to
support the TI LPC, it is at least a better starting ground than
reworking Andy Fingerhuts code. I'm wrapping Phil's api into a C# class
now and will run some sound wav's through the LPC encoder he has and
see what it generates through a TI chip (most likely it won't work). I
have all the TI lookup tables for the coefficients so I should be able
to extend Phil's code to output native TI speech packets along with the
TI enhancement of 'repeat' packets.

I have to admit, I really don't know anything about speech synthesis
but am coming at this for more of a specific need from left field and I
can code in C and C#. :-)

I will be sure to send anything I do to Phil in case he would like to
include it in HawkVoice in case some other obscure person might need
this functionality.

If anyone has any random suggestions or tips on working with the TI
LPC, let em fly, I will keep watch on this thread. You can also email
me directly through the contact link at any of the following
websites...

http://www.gamearchive.com
http://www.ipsnd.net
http://level7.askey.org

thanks... jess



Steve Underwood wrote:
> mhavoc wrote: > > Does anyone here possibly have detailed information on Texas > > Instruments implementation of the LPC algorithm for speech compression? > > I am in the process of writing an LPC-T1 and LPC-10 encoder/decoder > > library in C# and I have Andy Fingerhut's LPC-10 code (which is in C, > > converted from Fortran). From the data manuals I have, LPC-10 has 53 > > bits per frame plus a sync frame. TI LPC has 49 bits (no sync) plus a > > 'repeat' bit. The other bit differences are in the coefficient lookup > > tables sizes it appears. > > > > I have several of the TI TMS5XXX series datasheets on my website at... > > > > http://www.gamearchive.com/General/Data_Sheets/ > > > > Before I go into severe specifics, I just wanted to know if anyone here > > might be willing to help me with this topic or had more info.. > > A quick comparison shows not much in common, except for the basic scheme > of using LPC. The quantisation of every parameter is different. There > seems to be no special unvoiced processing in the TI LPC. > > Steve