Hi, I am trying to extend the voice_kal_diphone phoneset by introducing a single phone from Hindi and adding it to the phoneset. At this point, I am more interested in understanding the process of creating a new voice and (pending availability of data) am proceeding incrementally by adding a single phone to a complete voice. I am also recording the new words in my own voice and do not much care if the output suffers as a result of different voices (for different phones) in the output. My understanding of adding a new voice or phone can be separated into two parts:
1. Recording the wav files, labeling the input sentences and getting LPC coefficients for the phones. 2. Creating new lts rules for the new phone. I am trying to create new LTS rules for this phone using the instructions given here: http://festvox.org/bsv/x1469.html My question(s) are: 1. When I use merge_models on the aligned file, my LTS rules look like: ;; LTS rules (set! cmu_lts_rules '( )) This maybe because I am using only a very small training and test set (All tests until the creation of LTS rules have been successful). Why would this error arise and how can I get rid of it? 2. Is there a way that I can add to the LTS rules incrementally? Or do I need to recompile the lts_rules for the entire lexicon each time I enter a new phone? I apologize in case my question is unclear. I have just started to explore festival and some of my questions may come across as naive. Regards, Sriram
_______________________________________________ Festlang-talk mailing list [email protected] https://lists.berlios.de/mailman/listinfo/festlang-talk
