festvox / festival

Festival Speech Synthesis System
Other
376 stars 58 forks source link

cmudict bug ? #73

Open ddavout opened 3 months ago

ddavout commented 3 months ago

in cmudict we have

("certificate" n (((s er) 1) ((t ah) 1) ((f ax) 0) ((k ax t) 0)))
("certificate" v (((s er) 1) ((t ah) 1) ((f ax) 0) ((k ey t) 0)))

nevertheless

festival> (SayText "I certficate my certificate")

<Utterance 0x7f6a48fd0d90>

festival> (format nil "%l" (utt.relation.print ! (quote Word))))

()
id _5 ; name I ; pos nn ; pos_index 8 ; pos_index_score 0 ; phr_pos n ; phrase_score -5.64644 ; pbreak_index 1 ; pbreak_index_score 0 ; pbreak NB ;
id _6 ; name certficate ; pos_index 17 ; pos_index_score 0 ; pos vbd ; phr_pos v ; phrase_score -6.05754 ; pbreak_index 1 ; pbreak_index_score 0 ; pbreak NB ;
id _7 ; name my ; pos_index 12 ; pos_index_score 0 ; pos prp ; phr_pos prp ; phrase_score -5.4346 ; pbreak_index 1 ; pbreak_index_score 0 ; pbreak NB ;
id _8 ; name certificate ; pos_index 8 ; pos_index_score 0 ; pos nn ; phr_pos n ; pbreak_index 0 ; pbreak_index_score 0 ; pbreak B ; blevel 3 ;

we have festival> (utt.flat_repr !)

"<{(1 ay )}{(1 s er t )(0 f ax )(0 k ax t )}{(1 m ay )}{(1 s er )(1 t ah )(0 f ax )(0 k ax t )}> _B "

I was expecting

"<{(1 ay )}{(1 s er t )(0 f ax )(0 k ey t )}{(1 m ay )}{(1 s er )(1 t ah )(0 f ax )(0 k ax t )}> _B "

( of course I would prefer to hear the missing syllables (1 t ih ) but it is not a simple problem of bad pronunciation )