Hi,
If the units to be modeled have different
durations, such as the initial-final structure in mandarin Chinese,
maybe multiple proto type will be better, for example, 5 states for
initials and 7 states for finals. HTK supports this but in HTS, state duration
PDFs are represented in multi-stream structure, an error will be found
when making a monophone mmf for duration, what can I do if I want to inplement
this multiple proto type structure to the current HTS framework? Any
help?
Thank you
!
Best Regards,
Jingwei Sun
Jingwei
Sun. ThinkIT Speech Lab., Institute of Acoustics, CAS, China
2008-04-13
|