Hi Rob, Thank you very much for your comments. Unfortunately I have been asked to stick to the Unisyn synthesizer since that is the one we will be basing our voices etc. on.
I have been running festival with a debugger for the past so many days and have made a little progress. I am stuck in the function leval in slib.cc. I keep looping through this function and step through it as well but am not able to get to the underlying C++ code (where the LPC resynthesis occurs). I feel that this is the loop where the resynthesis happens. I have the following questions: 1. Which are the C++ functions where the LPC resynthesis takes place? 2. Where would the functions for unit selection be? Regards, Sriram Shankar. On Mon, Feb 6, 2012 at 3:51 AM, Rob Clark <[email protected]> wrote: > Probably the best way to see how LPC works is to download the following: > > The multisyn build tools. > http://www.cstr.ed.ac.uk/downloads/festival/multisyn_build/ > (These provide "unit selection" voice building tools, which include > scripts for creating LPC and Residuals) See the files in the > documentation directory. > > > http://www.cstr.ed.ac.uk/downloads/festival/1.95/festvox_cstr_us_awb_arctic_multisyn-1.0.tar.gz > > This is an example multisyn voice. It hasn't been maintained so won't > run unaltered with Festival 2.1 > (If you want to run it with 2.1, define the scheme function it > complains as missing as a function that does't do anything in the > scheme file which defines the voice.) > > Otherwise just look at the scheme voice definition in the above file > to see what scheme commands get called to do lpc synthesis. > Note that for these functions to work, they requires certain relations > in the Utt structure to exist and be meaningful. > > It is not too difficult to follow the scheme through to the C++. just > grep for things in src/modules/*. It is much harder to follow what the > signal processing code actually does and I haven't had to do so for a > few years, so I can't really provide much advice. > > It is also worth noting that the same code can just do overlap and add > if provided with a wave file and a pitch mark file rather than a > residual and lpc coefs. > > Multisyn doesn't do pitch modification by default, but it can be > turned on, it does sometimes break though with multisyn voice data, I > think mostly when you ask it to do modification by an unreasonable > amount. > > Regards, > Rob. > > On 5 February 2012 13:30, Nickolay V. Shmyrev <[email protected]> wrote: > > On 01/02/2012 at 13:54 +0530, Sriram Shankar wrote: > >> Hi, > >> > >> I am trying to go deeper into Festival by trying to understand the > >> code. As a task, I am trying to give a wave file from the command > >> line, create its LPC coefficients and then resynthesize with LPC. My > >> question: > >> > >> 1. What C++ classes can be used to analyze an input wave file into its > >> LPC coefficients? > > > > EST_sigpr_utt > > > >> 2. Is there a document that explains the underlying C++ classes in the > >> Festival code in greater detail? > > > > Yes, it's called "source code" > > > > > > -- > Rob Clark >
_______________________________________________ Festlang-talk mailing list [email protected] https://lists.berlios.de/mailman/listinfo/festlang-talk
