-
Notifications
You must be signed in to change notification settings - Fork 57
/
NEWS
219 lines (199 loc) · 8.34 KB
/
NEWS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
Note that not all features discussed in this file are included in
the standard distribution.
HISTORY
June 21st 2001 1.4.2 Release
Various new gcc's support
Visual C++ 6.0 support
uses configure (though could do so even more)
substantial updates to the clunits unit selection module
lots of wee bugs fixed
a few very hard bugs fixed
(client/server race condition)
(dropped bytes in reading files when machine overloaded)
(FreeBSD memory/gc problem)
default waverform type is now RIFF.
Nov 21st 1999 1.4.1 Release
SSFF (for emulabel) track support
AIX support
Java fixes
various minor bug fixes
WFST with proper quoting
Wagon sample counts
gcc-2.95.1 support
June 20th 1999 1.4.0 Release
becomes free software
June 6th 1999 1.3.95 Beta
size/speed/memory leak overhaul (no memory leaks)
XML support for relation loading (for SOLE support)
JSAPI initial support
GalaxyCommunicator architecture interface
ked_mttilt_diphone voice built
Parser trained on MARSEC (prosodic) brackets rather than syntax
Unisyn_selection fully integrated
Unisyn_phonology fully integrated
viterbi cart/ngram/wfst base LTS prediction (did improve but BIG)
viterbi cart/ngram based accent prediction (didn't improve)
tilt working (again)
audioin (na_record) for many architectures
viterbi from Scheme (with cart, ngram, wfst models)
January 26th 1999 1.3.1 Release
egcs-1.1.1 support
tobi_rules update (GM)
replace readline with editline (+ extensions)
Lots of little bug fixes
cluster code tidied up
kal voice
ked power normalization
updated lexicons with addenda for US and UK
New LTS models for US and UK English
"Building Voices in Festival" document
August 24th 1998 1.3.0 Release
UniSun/groupfile optimizations
Java client support
Fixed ESPS so both track and wave output works
Retraining of most modules with new architecture (durations improved)
rxp, (Richard's XML parser) integrated and Sable XMLified
Fringe display program for labels and utterances
Metrical tree synthesis
A new utterance architecture (Relations and Items)
utterance save and load work properly now
Trainable LTS system
Lexicon cache system
Substantial optimization of front end (twice the speed)
UniSyn, new signal processing and generic waveform synthesis module
OLS code added
WFST support for kk rules, regular grammars etc, simple English morphology
November 30th 1997 1.2.4 BETA
Tilt analysis and Tilt intonation modules added.
make_utts substantially improved (> 100 times faster)
text2wave script added
Pitch synchronous lpc analysis and support
rab consonant clusters labelled
New duration tree (wagon stepwise) much smaller if not better
SCFG grammar and parser (scfg_parse_text added as festival script)
change config stuff (again)
October 1st 1997 1.2.1 RELEASE
preliminary support for Visual C++
Use path-append rather than string-append (in buckets of places)
Minor bugs fixes throughout the code (end silences are now *always*
inserted in tts)
Linux socket bug fixed (get_url didn't work)
native irix audio support
September 5th 1997 1.2.0 RELEASE
Proclaim modules and voices
automatic detection of voices
Phonset, lexicon, ltsrules listing and printing
16 bit linear native support for Solaris i386 (sb16)
Update Festival Tutorial to 1.2.0
Aygust 15th 1997 1.1.99 Beta release
Win NT (and 95) initial support Cygnus win32 and Visual C++
100 more pages of documentation
LPC analysis for voices now ESPS independent
Spanish el voices tidy up (Borja)
ToBI by rule implementation
Confirmed support for gcc-2.7.2, gcc-2.6.3, Linux, FreeBSD, SunOS
Alpha and SunCC port on Solaris
reference card added
return s-expressions in server/client mode
OGI markup mode added.
Native support for sun16, linux16 and freebsd (compile time option)
Changed names of .C files to .cc files for bILL
wagon integrated into speech tools (plus docs)
auto-text-mode-alist for automatic selection of text mode from file name
Associated token tests added
Many more tokens dealt with (numbers, money, roman, phone, etc.)
(analysed databases to see what coverage is like)
A probablistic chart parser (no significant grammars though)
RJC's new database/units/join/modify modules taking shape
Some more examples added to the tutorial (with answers)
Integrated CSLU changes for OGItoolkit including TCL support
stml support for phrase types and words inline
ssml -> stml
Postlexical rules done in Scheme rather than C++
Rest of functions to allow any manipulation of utterance from scheme
New duration models trained for both English and American
New lexicon (CMU based)
Consonant cluster support (for kd)
American diphone set
Cluster unit selection algorithm more robust
Ngram backoff smoothing
Token pos, for numbers (97.5%) but does poor on phone numbers
New lexicon with final Rs and r deletion as postlex rule
Update pos prediction (ts39) and phrase break ngrams (faster to load)
New ngram format (binary files, and smoothing)
Vowel reduction module
Sun CC port
New string class (rjc) remove dependence on libg++
Update of course notes and new section on building models from dbs
Yarowsky homograph disambiguation
Jan 24th 1997 1.1.1 release (first public release)
a number of configuration and INSTALL documentation bugs fixed
SSML tidied up and a festival script provided for it.
Diphones, again, checked and copyright explicitly added
Jan 6th 1997 1.1.0 release
Roger diphones now default speaker
A new unit clustering algorithm with acoustic costs and
optimal coupling
BSD socket client/server support
A format function in Scheme (fprintf-like)
A short course on Speech Synthesis in Festival
(with course notes and exercises)
A programmable form of text modes including externally customizable
token to word rules.
Fully programmable intonation module (for ToBI-like theories)
Backtrace facility in Lisp
Externally specified Utterance end (for all tts modes) with lookahead
Roger diphones, first draft
Nov 8th 1.0.0 release
Substantial bug fixes, stabilization and documentation updates
Added residual excited LPC synthesizer and removed PSOLA code.
Made sucs and taylor optional modules, new modules
can be added without modifying the base code
MOBY lexicon (not as good as cuvoald but free)
New diphone grouping software
A new diphone database module (free from adc)
Sept 30th 0.1 release
MBROLA support (good example of external module)
latest news: read out the latest news (from Time Warner, Pathfinder)
audio spooler
--language option on command line
Spanish synthesis
Letter to sound rules as external system (replacing all the NRL code)
Welsh synthesis, making the whole system more language independent
sucs spoke in reasonably way (gsw_450 and f2b dbs)
document strings for functions (built in and user) and variables
access from command line and dumped automatically into texinfo
cleaned up SSML implementation
break prediction integrated using viterbi and pos
sucs module started (selection of units for concatenative synthesis)
a part-of-speech tagging system (ngram/viterbi based)
viterbi code added
fixes in SIOD for running batch and stdin, also sub_prompts added
saytime example
Memory leaks fixed, no leaks for tts
July 30th 0.0 release (just for the sake of it)
a significiant start at documentation (texinfo -> info & html)
festival scripts using #! on first line
donovan diphone support
can compile (with too many warnings) under g++ 2.7.2
copyrights on all files
memory leak checks (only 8 bytes for "unknown" words)
SSML (and tts file modes)
cuvoald cmu and beep lexicons
lexicon compilation
web page, emacs interface
June 2nd
Klatt duration module
syllabification in phones from letter to sound rules
Linear Regression model for F0 prediction (from ToBI labels)
CART (wagon) built trees for duration (zscores), phrase boundaries,
accent and endtone prediction.
ffeatures allowing specification of features of an utterance
May
integrated Taylor diphone module
US Naval Research letter to sound rules
CSTR lexicon
12th April first words "hello"
start with speech_tools library, scheme-in-one-defun and readline
and external CSTR diphone synthesizer
7th April 1996 work started