Hi, Thank you for your help. The following is how I resolved the problem:
I set the durations for the phone in the file cmu_us_kal_dur.scm. If I understand this correctly, these durations represent the average and std. deviation of the durations of "txh" (the phone I want to add) from across all the spoken prompts. I entered the same values as for a phone that was closest to "txh" and it worked. The result may not be awesome, but I have got it working. I will now try to automate this process of populating average and std. dev. of duration of the different phones using the programs listed in the link you mentioned. Thanks, Sriram. On Wed, Jan 11, 2012 at 7:32 PM, Nickolay V. Shmyrev <[email protected]>wrote: > You can set phone durations in festvox/*.scm files. You can find more > information in a book: > > http://festvox.org/bsv/x1900.html > > On Tue, Jan 10, 2012 at 12:03 PM, Sriram Shankar <[email protected]> > wrote: > > Hi, > > > > I am trying to add a new phone to the BU radio phoneset. The following > steps > > I have followed: > > > > 1. Generated the prompts: > > festival --heap 2100000 -b ../../festvox/src/diphones/diphlist.scm > > ../../festvox/src/diphones/us_withTXH_schema.scm '(diphone-gen-waves > > "prompt-wav" "prompt-lab" "etc/kaldiph_withTXH.list")' > > > > 2. Recorded the prompts > > > > 3. Labelled the prompts like so: > > ../festvox/src/general/do_build label_dtw > > > > 4. Getting the distribution of diphones using make_diph_index did not > work > > no matter what I tried. So I did this step manually by entering the > > following fields in the dic/kaldiph.est file: > > <diphone> <fileid> <start time> <mid time> <end time> (with all times in > > seconds). > > > > 5. Extracted pitch marks, lpc coefficients and mel cepstrum coefficients > > with the do_build function and appropriate arguments. > > > > 6. Also added some words containing this new phone to the lexicon, > trained > > new CART rules, compiled the new lexicon and set this up for use. > > > > When I run SayPhones from the command line, it seems to work OK. I can > hear > > the new phone added. It may not be very clear but it is audible. > > On running SayText (also from the command line), I get a message saying: > > Phoneme: txh has no duration info > > Using UniSyn method to synthesize an utterance. > > UniSyn: using default diphone pau-pau for s-txh > > > > Where am I going wrong? And how can I get txh to have a duration for > > SayText? > > > > Regards, > > Sriram Shankar. >
_______________________________________________ Festlang-talk mailing list [email protected] https://lists.berlios.de/mailman/listinfo/festlang-talk
