;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;; ;; ;;; Centre for Speech Technology Research ;; ;;; University of Edinburgh, UK ;; ;;; Copyright (c) 1997 ;; ;;; All Rights Reserved. ;; ;;; ;; ;;; Permission is hereby granted, free of charge, to use and distribute ;; ;;; this software and its documentation without restriction, including ;; ;;; without limitation the rights to use, copy, modify, merge, publish, ;; ;;; distribute, sublicense, and/or sell copies of this work, and to ;; ;;; permit persons to whom this work is furnished to do so, subject to ;; ;;; the following conditions: ;; ;;; 1. The code must retain the above copyright notice, this list of ;; ;;; conditions and the following disclaimer. ;; ;;; 2. Any modifications must be clearly marked as such. ;; ;;; 3. Original authors' names are not deleted. ;; ;;; 4. The authors' names are not used to endorse or promote products ;; ;;; derived from this software without specific prior written ;; ;;; permission. ;; ;;; ;; ;;; THE UNIVERSITY OF EDINBURGH AND THE CONTRIBUTORS TO THIS WORK ;; ;;; DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE, INCLUDING ;; ;;; ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO EVENT ;; ;;; SHALL THE UNIVERSITY OF EDINBURGH NOR THE CONTRIBUTORS BE LIABLE ;; ;;; FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES ;; ;;; WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN ;; ;;; AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ;; ;;; ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF ;; ;;; THIS SOFTWARE. ;; ;;; ;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;; ;;; Postlexical rules ;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;; Modifed for CSTR HTS Voice Library ;; ;; Author : Junichi Yamagishi (jyamagis@inf.ed.ac.uk) ;; ;; Date : Sept 2008 ;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; (define (PostLex utt) "(PostLex utt) Apply post lexical rules to segment stream. These may be almost arbitrary rules as specified by the particular voice, through the postlex_hooks variable. A number of standard post lexical rule sets are provided including reduction, posessives etc. These rules are also used to mark standard segments with their cluster information used in creating diphone names." (let ((rval (apply_method 'PostLex_Method utt))) (cond (rval rval) ;; new style (t ;; should only really need this one (apply_hooks postlex_rules_hooks utt))) utt )) (define (Classic_PostLex utt) "(Classic_PostLex utt) Apply post lexical rules (both builtin and those specified in postlex_rules_hooks)." (Builtin_PostLex utt) ;; haven't translated all the rules yet (apply_hooks postlex_rules_hooks utt) utt ) (defvar postlex_rules_hooks nil "postlex_rules_hooks A function or list of functions which encode post lexical rules. This will be voice specific, though some rules will be shared across languages.") ;;; Mapping of full vowels to reduced vowels, this should be part ;;; of the phoneset definitions (defvar postlex_vowel_reduce_table '((mrpa ((uh @) (i @) (a @) (e @) (u @) (o @) (oo @))) (radio ((ah ax el en em) (ih ax) ; (er axr ax) ; (iy ih) ; (ey ax) (aa ax) (ae ax) (eh ax)))) "postlex_vowel_reduce_table Mapping of vowels to their reduced form. This in an assoc list of phoneset name to an assoc list of full vowel to reduced form.") (defvar postlex_vowel_reduce_cart_tree nil "postlex_vowel_reduce_cart_tree CART tree for vowel reduction.") (defvar postlex_vowel_reduce_cart_tree_hand '((stress is 0) ((p.syl_break < 2) ((syl_break < 2) ((1)) ((0))) ((0))) ((0))) "postlex_vowel_reduce_cart_tree_hand A CART tree for vowel reduction. This is hand-written.") (defvar postlex_vowel_reduce_cart_data ' ((R:SylStructure.parent.gpos is cc) (((0 0.993548) (1 0.00645161) 0)) ((p.R:SylStructure.parent.gpos is md) (((0 0.903226) (1 0.0967742) 0)) ((p.R:SylStructure.parent.gpos is det) ((n.R:SylStructure.parent.gpos is content) ((last_accent < 2.5) ((next_accent < 2.5) ((next_accent < 1.2) ((n.syl_break is 4) (((0 0.967213) (1 0.0327869) 0)) ((syl_break is 4) (((0 0.952381) (1 0.047619) 0)) ((n.syl_break is 4) (((0 0.953488) (1 0.0465116) 0)) ((position_type is single) (((0 0.947368) (1 0.0526316) 0)) ((accented is 0) ((n.accented is 0) (((0 0.857143) (1 0.142857) 0)) (((0 0.415385) (1 0.584615) 1))) (((0 0.974359) (1 0.025641) 0))))))) (((0 0.968254) (1 0.031746) 0))) (((0 0.969697) (1 0.030303) 0))) (((0 0.976744) (1 0.0232558) 0))) (((0 0.990291) (1 0.00970874) 0))) ((next_accent < 108.5) ((p.R:SylStructure.parent.gpos is pps) (((0 0.828947) (1 0.171053) 0)) ((R:SylStructure.parent.gpos is det) ((accented is 0) (((0 0.0599572) (1 0.940043) 1)) (((0 0.949367) (1 0.0506329) 0))) ((p.R:SylStructure.parent.gpos is cc) (((0 0.880952) (1 0.119048) 0)) ((p.R:SylStructure.parent.gpos is wp) (((0 0.875) (1 0.125) 0)) ((p.R:SylStructure.parent.gpos is in) ((n.syl_break is 4) (((0 0.961538) (1 0.0384615) 0)) ((next_accent < 2.5) ((syl_break is 4) (((0 0.95122) (1 0.0487805) 0)) ((next_accent < 1.2) ((accented is 0) ((n.stress is 0) (((0 0.788462) (1 0.211538) 0)) ((R:SylStructure.parent.R:Word.p.gpos is content) (((0 0.863636) (1 0.136364) 0)) ((position_type is single) (((0 0.729167) (1 0.270833) 0)) (((0 0.4) (1 0.6) 1))))) (((0 0.983871) (1 0.016129) 0))) (((0 0.96) (1 0.04) 0)))) (((0 0.963636) (1 0.0363636) 0)))) ((position_type is single) ((syl_break is 4) (((0 0.993865) (1 0.00613497) 0)) ((p.R:SylStructure.parent.gpos is to) (((0 0.984375) (1 0.015625) 0)) ((syl_break is 1) ((accented is 0) ((n.R:SylStructure.parent.gpos is in) (((0 0.869565) (1 0.130435) 0)) ((R:SylStructure.parent.gpos is content) (((0 0.861789) (1 0.138211) 0)) ((p.R:SylStructure.parent.gpos is content) ((p.syl_break is 4) (((0 0.858065) (1 0.141935) 0)) ((R:SylStructure.parent.gpos is in) ((p.syl_break is 1) ((n.R:SylStructure.parent.gpos is det) (((0 0.659574) (1 0.340426) 0)) ((p.stress is 0) (((0 0.422222) (1 0.577778) 1)) (((0 0.582278) (1 0.417722) 0)))) ((n.accented is 0) ((n.R:SylStructure.parent.gpos is content) (((0 0.65) (1 0.35) 0)) ((p.stress is 0) (((0 0.464286) (1 0.535714) 1)) (((0 0.538462) (1 0.461538) 0)))) (((0 0.803279) (1 0.196721) 0)))) ((n.R:SylStructure.parent.gpos is det) (((0 0.952381) (1 0.047619) 0)) ((n.syl_break is 4) (((0 0.833333) (1 0.166667) 0)) ((p.stress is 0) ((p.syl_break is 1) ((n.syl_break is 1) (((0 0.740741) (1 0.259259) 0)) ((R:SylStructure.parent.gpos is aux) (((0 0.478261) (1 0.521739) 1)) (((0 0.769231) (1 0.230769) 0)))) (((0 0.755556) (1 0.244444) 0))) (((0 0.797619) (1 0.202381) 0))))))) (((0 0.870968) (1 0.129032) 0))))) (((0 0.983806) (1 0.0161943) 0))) (((0 0.977778) (1 0.0222222) 0))))) ((next_accent < 21.6) ((p.stress is 0) ((R:SylStructure.parent.R:Word.p.gpos is md) (((0 0.961538) (1 0.0384615) 0)) ((position_type is mid) (((0 0.977612) (1 0.0223881) 0)) ((n.R:SylStructure.parent.gpos is det) (((0 0.916667) (1 0.0833333) 0)) ((R:SylStructure.parent.R:Word.n.gpos is 0) (((0 0.915493) (1 0.084507) 0)) ((R:SylStructure.parent.R:Word.n.gpos is pps) (((0 0.884615) (1 0.115385) 0)) ((n.stress is 0) ((n.syl_break is 4) (((0 0.986755) (1 0.013245) 0)) ((p.syl_break is 4) (((0 0.977011) (1 0.0229885) 0)) ((n.syl_break is 4) (((0 0.965517) (1 0.0344828) 0)) ((last_accent < 1.2) ((last_accent < 0.1) (((0 0.910448) (1 0.0895522) 0)) ((next_accent < 1.2) ((R:SylStructure.parent.R:Word.n.gpos is in) (((0 0.82) (1 0.18) 0)) ((n.syl_break is 0) ((R:SylStructure.parent.R:Word.p.gpos is content) (((0 0.819672) (1 0.180328) 0)) (((0 0.444444) (1 0.555556) 1))) (((0 0.785714) (1 0.214286) 0)))) (((0 0.836364) (1 0.163636) 0)))) (((0 0.962025) (1 0.0379747) 0)))))) ((stress is 0) ((n.syl_break is 4) (((0 0.21875) (1 0.78125) 1)) ((R:SylStructure.parent.R:Word.p.gpos is aux) (((0 0.259259) (1 0.740741) 1)) ((p.syl_break is 1) (((0 0.243094) (1 0.756906) 1)) ((R:SylStructure.parent.R:Word.p.gpos is det) (((0 0.290323) (1 0.709677) 1)) ((R:SylStructure.parent.R:Word.p.gpos is in) (((0 0.3) (1 0.7) 1)) ((syl_break is 1) (((0 0.289157) (1 0.710843) 1)) ((p.syl_break is 4) (((0 0.352941) (1 0.647059) 1)) ((n.syl_break is 0) (((0 0.311475) (1 0.688525) 1)) ((syl_break is 4) (((0 0.4) (1 0.6) 1)) (((0 0.581395) (1 0.418605) 0))))))))))) (((0 1) (1 0) 0))))))))) ((stress is 0) ((R:SylStructure.parent.R:Word.n.gpos is 0) (((0 0.121212) (1 0.878788) 1)) ((next_accent < 2.4) ((R:SylStructure.parent.gpos is content) ((position_type is mid) (((0 0.176895) (1 0.823105) 1)) ((p.syl_break is 1) (((0 0.229167) (1 0.770833) 1)) ((syl_break is 4) (((0 0.242775) (1 0.757225) 1)) ((p.syl_break is 0) ((n.R:SylStructure.parent.gpos is in) (((0 0.253521) (1 0.746479) 1)) ((R:SylStructure.parent.R:Word.p.gpos is in) (((0 0.262774) (1 0.737226) 1)) ((last_accent < 2.1) ((n.R:SylStructure.parent.gpos is aux) (((0 0.304348) (1 0.695652) 1)) ((next_accent < 1.2) ((n.R:SylStructure.parent.gpos is cc) (((0 0.291667) (1 0.708333) 1)) ((syl_break is 1) ((n.syl_break is 4) (((0 0.344828) (1 0.655172) 1)) ((R:SylStructure.parent.R:Word.p.gpos is det) (((0 0.364706) (1 0.635294) 1)) ((n.syl_break is 4) (((0 0.384615) (1 0.615385) 1)) ((last_accent < 1.2) ((p.accented is 0) (((0 0.584906) (1 0.415094) 0)) ((n.accented is 0) ((R:SylStructure.parent.R:Word.p.gpos is content) (((0 0.41) (1 0.59) 1)) (((0 0.6) (1 0.4) 0))) (((0 0.333333) (1 0.666667) 1)))) (((0 0.380952) (1 0.619048) 1)))))) ((p.accented is 0) (((0 0.183673) (1 0.816327) 1)) ((n.R:SylStructure.parent.gpos is content) ((n.stress is 0) (((0 0.295455) (1 0.704545) 1)) ((R:SylStructure.parent.R:Word.p.gpos is content) ((n.syl_break is 1) (((0 0.5) (1 0.5) 0)) (((0 0.40625) (1 0.59375) 1))) (((0 0.333333) (1 0.666667) 1)))) (((0 0.2) (1 0.8) 1)))))) (((0 0.3) (1 0.7) 1)))) (((0 0.302326) (1 0.697674) 1))))) (((0 0.25) (1 0.75) 1)))))) (((0 0.173913) (1 0.826087) 1))) (((0 0.166667) (1 0.833333) 1)))) (((0 1) (1 0) 0)))) (((0 0.2) (1 0.8) 1))))))))) (((0 0.15) (1 0.85) 1))))))) (defvar postlex_mrpa_r_cart_tree '((name is r) ((R:Segment.n.ph_vc is -) ((delete)) ((nil))) ((nil))) "postlex_mrpa_r_cart_tree For remove final R when not between vowels.") ;; Changed this to actually work... (Rob 09/12/04) ;; Changed this to delete the syllable when schwa is unneccesary (awb 19/07/04) (define (postlex_apos_s_check utt) "(postlex_apos_s_check UTT) Deal with possesive s for English (American and British). Delete schwa of 's if previous is not an alveolar or palatal fricative or affricative, and change voiced to unvoiced s if previous is not voiced." (mapcar (lambda (syl) ; word is 's (if (string-equal "'s" (item.feat syl "R:SylStructure.parent.name")) (begin ;; de-voice if last phone of previous word is unvoiced (if (string-equal "-" (item.feat syl "p.R:SylStructure.daughtern.ph_cvox")) (item.set_name (item.relation.daughtern syl 'SylStructure) "s")) ;; change it from "z" to "s" ; if the previous seg is a aveolar or palatal, ; fricative or affricate don't delete schwa otherwise delete it (if (and (member_string (item.feat syl "p.R:SylStructure.daughtern.ph_ctype") '(f a)) (member_string (item.feat syl "p.R:SylStructure.daughtern.ph_cplace") '(a p))) (begin t) (begin ;; delete the schwa (item.delete (item.relation.daughter1 syl 'SylStructure)) ;; attach orphaned s/z to previous word (item.relation.append_daughter (item.prev syl) 'SylStructure (item.relation.daughtern syl 'SylStructure)) ;; delete the now empty syllable (item.delete syl)))))) ;; never happens to if 's is first in an utterance (cdr (utt.relation.items utt 'Syllable))) utt) ;; Changed this to work the other way round, too. Volker 10/08/06 (define (postlex_the_vs_thee utt) "(postlex_the_vs_thee utt) Unnreduce the schwa in \"the\" when a vowel follows. Reduce the vowel in \"the\" when no vowel follows (this requires a lexicon entry for \"the\" with feature \"reduced\", otherwise there will be no reduction)." (let ((fullform (cadr (car (caar (cdr (cdar (lex.lookup_all 'thee))))))) (reducedform (cadr(car(caar(cddr(lex.lookup 'the '(reduced))))))) seg) (mapcar (lambda (word) (if (string-equal "the" (downcase (item.feat word "name"))) (begin (set! seg (item.relation (item.daughtern (item.relation.daughtern word 'SylStructure)) 'Segment)) (if (string-equal "+" (item.feat (item.next seg) 'ph_vc)) (item.set_feat seg 'name fullform) (item.set_feat seg 'name reducedform))))) (utt.relation.items utt 'Word))) utt) (define (postlex_the_vs_thee_changeflag utt) "(postlex_the_vs_thee_changeflag utt) Unnreduce the schwa in \"the\" when a vowel follows. Reduce the vowel in \"the\" when no vowel follows (this requires a lexicon entry for \"the\" with feature \"reduced\", otherwise there will be no reduction)." (let ((fullform (cadr (car (caar (cdr (cdar (lex.lookup_all 'thee))))))) (reducedform (cadr(car(caar(cddr(lex.lookup 'the '(reduced))))))) seg) (mapcar (lambda (word) (if (string-equal "the" (downcase (item.feat word "name"))) (begin (set! seg (item.relation (item.daughtern (item.relation.daughtern word 'SylStructure)) 'Segment)) (if (string-equal "+" (item.feat (item.next seg) 'ph_vc)) (item.set_feat seg 'reducable 0) (item.set_feat seg 'reducable 1))))) (utt.relation.items utt 'Word))) utt) ;; For Multisyn voices only. Volker 14/08/06 (define (postlex_a utt) "(postlex_a utt) If POS of \"a\" is \"nn\" and segment feature \"reducable\", set it to 0. This is a bugfix, but still requires the target cost function to add a penalty if a candidate is reducable but the target is not. expro_target_cost does that." (let(seg) (mapcar (lambda(word) ;; (format t "%s\t%s\n" (item.feat word 'name)(item.feat word 'pos)) (if(and(string-equal "a" (downcase (item.feat word "name"))) (string-equal "nn" (item.feat word "pos"))) (begin (set! seg (item.relation (item.daughtern (item.relation.daughtern word 'SylStructure)) 'Segment)) ;; (format t "should not be reducable\n") (if (eq 1 (parse-number (item.feat seg 'reducable))) (item.set_feat seg 'reducable 0)))) ) (utt.relation.items utt 'Word))) utt) (define (postlex_unilex_vowel_reduction utt) "(postlex_unilex_vowel_reduction utt) Perform vowel reduction based on unilex specification of what can be reduced." (let () (mapcar (lambda (seg) (if (and (eq? (parse-number (item.feat seg "reducable")) 1) (not (> (parse-number (item.feat seg "R:SylStructure.parent.stress")) 0))) (if (not (and (seg_word_final seg) (string-equal (item.feat (item.next seg) 'ph_vc) "+"))) (item.set_feat seg "name" (item.feat seg "reducedform"))))) (utt.relation.items utt 'Segment))) utt) (define (seg_word_final seg) "(seg_word_final seg) Is this segment word final?" (let ((this_seg_word (item.parent (item.relation.parent seg 'SylStructure))) (silence (car (cadr (car (PhoneSet.description '(silences)))))) next_seg_word) (if (item.next seg) (set! next_seg_word (item.parent (item.relation.parent (item.next seg) 'SylStructure)))) (if (or (equal? this_seg_word next_seg_word) (string-equal (item.feat seg "name") silence)) nil t))) ;; imported from postlex_intervoc_r.scm Volker 14/08/06 (define (postlex_intervoc_r utt) "(postlex_intervoc_r UTT) Remove any word-final /r/ which is phrase-final or not going to be inter-vocalic i.e. the following words does not start with a vowel. NOTE: in older versions of unilex-rpx.out for Festival, there is no word-final /r/. " (let (word next_word last_phone following_phone) (set! word (utt.relation.first utt 'Word)) (while word (set! next_word (item.next word)) (set! last_phone (item.daughtern (item.daughtern(item.relation word 'SylStructure)))) (if next_word (begin (set! following_phone (item.daughter1 (item.daughter1 (item.relation next_word 'SylStructure)))) ; last_phone and following_phone should always be defined at this point, ; but since the upgrade to Fedora and characters no longer being in ISO ; but in UTF8, the pound sterling is no longer treated correctly. ; Probably (Token utt) should be fixed. (if (and following_phone last_phone) (begin (format t "%s\t%s %s %s %s\n" (item.name word) (item.name last_phone) (item.name following_phone) (item.feat following_phone 'ph_vc) (item.feat word 'pbreak)) (if(and(equal? "r" (item.name last_phone)) (or(not(equal? "NB" (item.feat word 'pbreak))) (equal? "-" (item.feat following_phone 'ph_vc)))) (begin (format t "\t\t\t/r/ in \"%s %s\" deleted\n" (item.name word)(item.name next_word)) (item.delete last_phone)))))) (if(and last_phone (equal? "r" (item.name last_phone))) (begin (format t "\t\t\tutterance-final /r/ deleted\n") (item.delete last_phone))) ) (set! word (item.next word)))) utt) (define (postlex_stop_deletion utt) "(postlex_stop_deletion utt) Delete any stop or affricative (phone which has a closure) immediately followed by another stop or affricative. Also save the identity of the deleted phone for the context cost functions. Consider: backtrack /b a k t r a k/ -> /b a t r a k/ (actually Jenny reduces : /b a k_cl k t_cl t r a k/ -> /b a k_cl t r a k/) If we then look for a diphone /a t/ we want to favour candidates coming from the same context i.e. which are actually a reduced /a k t/. In the data base, the 1st /a/ gets the feature right_context=k and the /t/ gets the fearture left_context=k. " (let(seg next_seg prev_seg) (set! seg (utt.relation.first utt 'Segment)) (while seg (set! prev_seg (item.prev seg)) (if prev_seg (begin ;(format t "%s %s %s\n" (item.name seg) ; (item.feat seg 'ph_ctype) ; (item.feat seg 'p.ph_ctype)) (if(and(or(equal? "s" (item.feat seg 'ph_ctype)) (equal? "a" (item.feat seg 'ph_ctype))) (or(equal? "s" (item.feat seg 'p.ph_ctype)) (equal? "a" (item.feat seg 'p.ph_ctype))) ; When there are 3 stops in a row, and after the 1st has been ; deleted, this prevents the 2nd to be deleted as well: (equal? 0 (item.feat prev_seg 'left_context))) (begin (set! prev_prev_seg (item.prev prev_seg)) (format t "postlex_stop_deletion: %s in %s\n" (item.name prev_seg) (item.name(item.parent(item.relation.parent prev_seg 'SylStructure)))) (if prev_prev_seg (begin ;(format t "setting left_context of %s and right context of %s to %s\n" ; (item.name seg) ; (item.name prev_prev_seg) ; (item.name prev_seg)) (item.set_feat seg 'left_context (item.name prev_seg)) (item.set_feat prev_prev_seg 'right_context (item.name prev_seg)))) (if(and(item.next seg) (equal? (item.name seg) (item.name prev_seg))) (begin ;(format t "setting left_context of %s to %s\n" ; (item.name (item.next seg) ; (item.name prev_seg)) (item.set_feat (item.next seg) 'left_context (item.name prev_seg)))) (item.delete prev_seg))))) (set! seg (item.next seg)))) utt) (provide 'postlex)