I would like to suggest a major advance in architecture. I have already
discussed it with nshm, who says that (a) it is sensible, (b) it is novel, (c)
it is a ton of work.
Currently Sphinx works simply from the order of phonemes it detects.
so ' engines turn ' can easily get recognised as ' engine is turn '.
If Sphinx were also to count syllables, it could throw out this
possibility:
'engines turn' has two syllables, whereas ' engine is turn ' has three
A further tuning would be to monitor the syllable stress for each syllable
S_n and record the ratio S_(k-1) : S_k and S_k : S_(k+1)
For example ' behave ' has the word stress on the second syllable, whereas ' behemoth ' has the stress on the first, and ' Bee-Hive ' has fairly even
weighting between the syllables.
I lay this at the feet of the Sphinx.
Ohmu
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Is this release roadmapped yet?
I would like to suggest a major advance in architecture. I have already
discussed it with nshm, who says that (a) it is sensible, (b) it is novel, (c)
it is a ton of work.
Currently Sphinx works simply from the order of phonemes it detects.
so ' engines turn ' can easily get recognised as ' engine is turn '.
If Sphinx were also to count syllables, it could throw out this
possibility:
'engines turn' has two syllables, whereas ' engine is turn ' has
three
A further tuning would be to monitor the syllable stress for each syllable
S_n and record the ratio S_(k-1) : S_k and S_k : S_(k+1)
For example ' behave ' has the word stress on the second syllable, whereas '
behemoth ' has the stress on the first, and ' Bee-Hive ' has fairly even
weighting between the syllables.
I lay this at the feet of the Sphinx.
Ohmu