Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss
HMM alignment, phoneme boundaries..
31 views
Skip to first unread message
Mauritz Jameson
Jan 20, 2012, 9:12:48 AM1/20/12
to
What is the most accurate algorithm for detecting and estimating
phoneme boundaries?
I am not looking for forced alignment algorithms where you give the
aligner the text or phoneme sequence because those algorithms usually
just gives an approximation of the phoneme boundaries (timing accuracy
too bad for building di-phone database).
What is the best accuracy current HMM aligners can give, and which one
is best?
Some HMM ASR can do "phoneme recognition" (without knowing phonemes
ahead of time), but I don't
know how accurate these can be. I suspect these wouldn't be accurate
enough for automatic
diphone creation when the input is an un-transcribed audio track - but
there are ways to train
on such data in "un-supervised" mode. But I don't know much about
this.
phoneme boundaries?
I am not looking for forced alignment algorithms where you give the
aligner the text or phoneme sequence because those algorithms usually
just gives an approximation of the phoneme boundaries (timing accuracy
too bad for building di-phone database).
What is the best accuracy current HMM aligners can give, and which one
is best?
Some HMM ASR can do "phoneme recognition" (without knowing phonemes
ahead of time), but I don't
know how accurate these can be. I suspect these wouldn't be accurate
enough for automatic
diphone creation when the input is an un-transcribed audio track - but
there are ways to train
on such data in "un-supervised" mode. But I don't know much about
this.
0 new messages