how to realize sequence of word occurences from word lattices and occurence of another word that is detected in phoneme lattices in lattice of one utterance with kaldi?
can I combine word and phoneme lattices with KALDI?
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Your question is a little bit hard to follow, but you can build simple
transducers and use composition
to perform the detection of certain sequences.
A simple Fst to detect word sequences would look something like this
(OpenFst text format):
0 0 Sigma <eps>
0 1 Word1 Word1
1 2 Word2 Word2
2 2 Sigma <eps>
2
Then apply it using fstcompose(lattice, detector)
The Sigma will match the entire vocabulary in the lattice, you can either
add an arc for every word
or use the advanced matchers in OpenFst.
The <eps> will map all others labels on a Word1,Word2 path to null.
Any path that does now have the sequence Word1,Word2 will be removed. You
will need a more complicated
machine if you need to detect the same sequence multiple times on the same
path.
how to realize sequence of word occurences from word lattices and
occurence of another word that is detected in phoneme lattices in lattice
of one utterance with kaldi?
can I combine word and phoneme lattices with KALDI?
how to realize sequence of word occurences from word lattices and occurence of another word that is detected in phoneme lattices in lattice of one utterance with kaldi?
can I combine word and phoneme lattices with KALDI?
Your question is a little bit hard to follow, but you can build simple
transducers and use composition
to perform the detection of certain sequences.
A simple Fst to detect word sequences would look something like this
(OpenFst text format):
0 0 Sigma <eps>
0 1 Word1 Word1
1 2 Word2 Word2
2 2 Sigma <eps>
2
Then apply it using fstcompose(lattice, detector)
The Sigma will match the entire vocabulary in the lattice, you can either
add an arc for every word
or use the advanced matchers in OpenFst.
The <eps> will map all others labels on a Word1,Word2 path to null.
Any path that does now have the sequence Word1,Word2 will be removed. You
will need a more complicated
machine if you need to detect the same sequence multiple times on the same
path.
On 18 April 2014 13:42, marzieh ghadirinia ghadirinia@users.sf.net wrote:
thanks..:)