tools/voice.pl

   1 #!/usr/bin/perl -s
   2 #             __________               __   ___.
   3 #   Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
   4 #   Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
   5 #   Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
   6 #   Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
   7 #                     \/            \/     \/    \/            \/
   8 # $Id$
   9 #
  10 # Copyright (C) 2007 Jonas Häggqvist
  11 #
  12 # All files in this archive are subject to the GNU General Public License.
  13 # See the file COPYING in the source tree root for full license agreement.
  14 #
  15 # This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
  16 # KIND, either express or implied.
  17
  18 use strict;
  19 use warnings;
  20 use File::Basename;
  21 use File::Copy;
  22 use Switch;
  23 use vars qw($V $C $t $l $e $E $s $S $i $v);
  24 use IPC::Open2;
  25 use IPC::Open3;
  26 use Digest::MD5 qw(md5_hex);
  27 use DirHandle;
  28 use open ':encoding(utf8)';
  29 use open ':std';
  30
  31 sub printusage {
  32     print <<USAGE
  33
  34 Usage: voice.pl [options] [path to dir]
  35  -V
  36     Create voice file. You must also specify -t and -l.
  37
  38  -C
  39     Create .talk clips.
  40
  41  -t=<target>
  42     Specify which target you want to build voicefile for. Must include
  43     any features that target supports.
  44
  45  -i=<target_id>
  46     Numeric target id. Needed for voice building.
  47
  48  -l=<language>
  49     Specify which language you want to build. Without .lang extension.
  50
  51  -e=<encoder>
  52     Which encoder to use for voice strings
  53
  54  -E=<encoder options>
  55     Which encoder options to use when compressing voice strings. Enclose
  56     in double quotes if the options include spaces.
  57
  58  -s=<TTS engine>
  59     Which TTS engine to use.
  60
  61  -S=<TTS engine options>
  62     Options to pass to the TTS engine. Enclose in double quotes if the
  63     options include spaces.
  64
  65  -v
  66     Be verbose
  67 USAGE
  68 ;
  69 }
  70
  71 # Initialize TTS engine. May return an object or value which will be passed
  72 # to voicestring and shutdown_tts
  73 sub init_tts {
  74     our $verbose;
  75     my ($tts_engine, $tts_engine_opts, $language) = @_;
  76     my %ret = ("name" => $tts_engine);
  77     switch($tts_engine) {
  78         case "festival" {
  79             print("> festival $tts_engine_opts --server\n") if $verbose;
  80             my $pid = open(FESTIVAL_SERVER, "| festival $tts_engine_opts --server > /dev/null 2>&1");
  81             my $dummy = *FESTIVAL_SERVER; #suppress warning
  82             $SIG{INT} = sub { kill TERM => $pid; print("foo"); panic_cleanup(); };
  83             $SIG{KILL} = sub { kill TERM => $pid; print("boo"); panic_cleanup(); };
  84             $ret{"pid"} = $pid;
  85         }
  86         case "sapi" {
  87             my $toolsdir = dirname($0);
  88             my $path = `cygpath $toolsdir -a -w`;
  89             chomp($path);
  90             $path = $path . '\\';
  91             my $cmd = $path . "sapi_voice.vbs /language:$language $tts_engine_opts";
  92             $cmd =~ s/\\/\\\\/g;
  93             print("> cscript //nologo $cmd\n") if $verbose;
  94             my $pid = open2(*CMD_OUT, *CMD_IN, "cscript //nologo $cmd");
  95             binmode(*CMD_IN, ':encoding(utf16le)');
  96             binmode(*CMD_OUT, ':encoding(utf16le)');
  97             $SIG{INT} = sub { print(CMD_IN "QUIT\r\n"); panic_cleanup(); };
  98             $SIG{KILL} = sub { print(CMD_IN "QUIT\r\n"); panic_cleanup(); };
  99             print(CMD_IN "QUERY\tVENDOR\r\n");
 100             my $vendor = readline(*CMD_OUT);
 101             $vendor =~ s/\r\n//;
 102             %ret = (%ret,
 103                     "stdin" => *CMD_IN,
 104                     "stdout" => *CMD_OUT,
 105                     "vendor" => $vendor);
 106         }
 107     }
 108     return \%ret;
 109 }
 110
 111 # Shutdown TTS engine if necessary.
 112 sub shutdown_tts {
 113     my ($tts_object) = @_;
 114     switch($$tts_object{"name"}) {
 115         case "festival" {
 116             # Send SIGTERM to festival server
 117             kill TERM => $$tts_object{"pid"};
 118         }
 119         case "sapi" {
 120             print({$$tts_object{"stdin"}} "QUIT\r\n");
 121             close($$tts_object{"stdin"});
 122         }
 123     }
 124 }
 125
 126 # Apply corrections to a voice-string to make it sound better
 127 sub correct_string {
 128     our $verbose;
 129     my ($string, $language, $tts_object) = @_;
 130     my $orig = $string;
 131     switch($language) {
 132         # General for all engines and languages
 133         $string =~ s/USB/U S B/g;
 134         $string =~ s/ID3/I D 3/g;
 135
 136         case "english" {
 137             switch($$tts_object{"name"}) {
 138                 case ["sapi","festival"] {
 139                     $string =~ s/plugin(s?)/plug-in$1/ig; next
 140                 }
 141                 case "festival" {
 142                     $string =~ s/\ba\b/ay/ig;
 143                     $string =~ s/$/./;
 144                 }
 145             }
 146         }
 147         case "deutsch" {
 148             # for all german engines (e.g. for english words)
 149             $string =~ s/alkaline/alkalein/ig;
 150             $string =~ s/byte(s?)/beit$1/ig;
 151             $string =~ s/clip(s?)/klipp$1/ig;
 152             $string =~ s/\bcover/kawwer/ig;
 153             $string =~ s/cuesheet/kjuschiet/ig;
 154             $string =~ s/dither/didder/ig;
 155             $string =~ s/equalizer/iquileiser/ig;
 156             $string =~ s/\bflash\b/fläsh/ig;
 157             $string =~ s/\bfirmware(s?)\b/firmwer$1/ig;
 158             $string =~ s/\bI D 3 tag\b/I D 3 täg/ig; # can't just use "tag" here
 159             $string =~ s/\bloudness\b/laudness/ig;
 160             $string =~ s/\bunicode\b/unikod/ig;
 161             switch($$tts_object{"name"}) {
 162                  case "sapi" {   # just for SAPI
 163                     switch($$tts_object{"vendor"}) {
 164                         case "AT&T Labs" {
 165                             $string =~ s/alphabet/alfabet/ig;
 166                             $string =~ s/ampere/amper/ig;
 167                             $string =~ s/\bdezibel\b/de-zibell/ig;
 168                             $string =~ s/diddering/didde-ring/ig;
 169                             $string =~ s/energie\b/ener-gie/ig;
 170                             $string =~ s/\Blauf\b/-lauf/ig;
 171                             $string =~ s/\bnumerisch\b/numehrisch/ig;
 172                         }
 173                     }
 174                 }
 175             }
 176         }
 177         case "svenska" {
 178             # for all swedish engines (e.g. for english words)
 179             $string =~ s/kilobyte/kilobajt/ig;
 180             $string =~ s/megabyte/megabajt/ig;
 181             $string =~ s/gigabyte/gigabajt/ig;
 182             $string =~ s/\bloudness\b/laudness/ig;
 183
 184             switch($$tts_object{"name"}) {
 185                  case "espeak" {   # just for eSpeak
 186                      $string =~ s/ampere/ampär/ig;
 187                      $string =~ s/bokmärken/bok-märken/ig;
 188                      $string =~ s/generella/schenerella/ig;
 189                      $string =~ s/dithering/diddering/ig;
 190                      $string =~ s/\bunicode\b/jynikod/ig;
 191                      $string =~ s/uttoning/utoning/ig;
 192                      $string =~ s/procent/pro-cent/ig;
 193                      $string =~ s/spellistor/spelistor/ig;
 194                      $string =~ s/cuesheet/qjyschiit/ig;
 195                  }
 196             }
 197         }
 198         case "italiano" {
 199             # for all italian engines (e.g. for english words)
 200             $string =~ s/Replaygain/Ripleyghein/ig;
 201             $string =~ s/Crossfade/Crossfeid/ig;
 202             $string =~ s/beep/Bip/ig;
 203             $string =~ s/cuesheet/chiushit/ig;
 204             $string =~ s/fade/feid/ig;
 205             $string =~ s/Crossfeed/crossfid/ig;
 206             $string =~ s/Cache/chash/ig;
 207             $string =~ s/\bfirmware(s?)\b/firmuer$1/ig;
 208             $string =~ s/\bFile(s?)\b/fail$1/ig;
 209             $string =~ s/\bloudness\b/laudness/ig;
 210             $string =~ s/\bunicode\b/unikod/ig;
 211             $string =~ s/Playlist/pleylist/ig;
 212             $string =~ s/WavPack/wave pak/ig;
 213             $string =~ s/BITRATE/bit reit/ig;
 214             $string =~ s/Codepage/cod page/ig;
 215             $string =~ s/PCM Wave/pcm Ue'iv/ig;
 216             switch($$tts_object{"name"}) {
 217                  case "sapi" {   # just for SAPI
 218                     switch($$tts_object{"vendor"}) {
 219                         case "Loquendo" {
 220                             $string =~ s/Inizializza/inizializa/ig;
 221                         }
 222                         case "ScanSoft, Inc" {
 223                             $string =~ s/V/v/ig;
 224                             $string =~ s/X/x/ig;
 225                             $string =~ s/stop/stohp/ig;
 226                         }
 227                     }
 228                 }
 229             }
 230         }
 231     }
 232     if ($orig ne $string) {
 233         printf("%s -> %s\n", $orig, $string) if $verbose;
 234     }
 235     return $string;
 236 }
 237
 238 # Produce a wav file of the text given
 239 sub voicestring {
 240     our $verbose;
 241     my ($string, $output, $tts_engine_opts, $tts_object) = @_;
 242     my $cmd;
 243     printf("Generate \"%s\" with %s in file %s\n", $string, $$tts_object{"name"}, $output) if $verbose;
 244     switch($$tts_object{"name"}) {
 245         case "festival" {
 246             # festival_client lies to us, so we have to do awful soul-eating
 247             # work with IPC::open3()
 248             $cmd = "festival_client --server localhost --otype riff --ttw --output \"$output\"";
 249             # Use festival-prolog.scm if it's there (created by user of tools/configure)
 250             if (-f "festival-prolog.scm") {
 251                 $cmd .= " --prolog festival-prolog.scm";
 252             }
 253             print("> $cmd\n") if $verbose;
 254             # Open command, and filehandles for STDIN, STDOUT, STDERR
 255             my $pid = open3(*CMD_IN, *CMD_OUT, *CMD_ERR, $cmd);
 256             # Put the string to speak into STDIN and close it
 257             print(CMD_IN $string);
 258             close(CMD_IN);
 259             # Read all output from festival_client (because it LIES TO US)
 260             while (<CMD_ERR>) {
 261             }
 262             close(CMD_OUT);
 263             close(CMD_ERR);
 264         }
 265         case "flite" {
 266             $cmd = "flite $tts_engine_opts -t \"$string\" \"$output\"";
 267             print("> $cmd\n") if $verbose;
 268             `$cmd`;
 269         }
 270         case "espeak" {
 271             $cmd = "espeak $tts_engine_opts -w \"$output\"";
 272             print("> $cmd\n") if $verbose;
 273             open(ESPEAK, "| $cmd");
 274             print ESPEAK $string . "\n";
 275             close(ESPEAK);
 276         }
 277         case "sapi" {
 278             print({$$tts_object{"stdin"}} "SPEAK\t$output\t$string\r\n");
 279         }
 280         case "swift" {
 281             $cmd = "swift $tts_engine_opts -o \"$output\" \"$string\"";
 282             print("> $cmd\n") if $verbose;
 283             system($cmd);
 284         }
 285     }
 286 }
 287
 288 # trim leading / trailing silence from the clip
 289 sub wavtrim {
 290     our $verbose;
 291     my ($file, $threshold, $tts_object) = @_;
 292     printf("Trim \"%s\"\n", $file) if $verbose;
 293     my $cmd = "wavtrim \"$file\" $threshold";
 294     if ($$tts_object{"name"} eq "sapi") {
 295         print({$$tts_object{"stdin"}} "EXEC\t$cmd\r\n");
 296     }
 297     else {
 298         print("> $cmd\n") if $verbose;
 299         `$cmd`;
 300     }
 301 }
 302
 303 # Encode a wav file into the given destination file
 304 sub encodewav {
 305     our $verbose;
 306     my ($input, $output, $encoder, $encoder_opts, $tts_object) = @_;
 307     printf("Encode \"%s\" with %s in file %s\n", $input, $encoder, $output) if $verbose;
 308     my $cmd = "$encoder $encoder_opts \"$input\" \"$output\"";
 309     if ($$tts_object{"name"} eq "sapi") {
 310         print({$$tts_object{"stdin"}} "EXEC\t$cmd\r\n");
 311     }
 312     else {
 313         print("> $cmd\n") if $verbose;
 314         `$cmd`;
 315     }
 316 }
 317
 318 # synchronize the clip generation / processing if it's running in another process
 319 sub synchronize {
 320     my ($tts_object) = @_;
 321     if ($$tts_object{"name"} eq "sapi") {
 322         print({$$tts_object{"stdin"}} "SYNC\t42\r\n");
 323         my $wait = readline($$tts_object{"stdout"});
 324         #ignore what's actually returned
 325     }
 326 }
 327
 328 # Run genlang and create voice clips for each string
 329 sub generateclips {
 330     our $verbose;
 331     my ($language, $target, $encoder, $encoder_opts, $tts_engine, $tts_engine_opts) = @_;
 332     my $english = dirname($0) . '/../apps/lang/english.lang';
 333     my $langfile = dirname($0) . '/../apps/lang/' . $language . '.lang';
 334     my $id = '';
 335     my $voice = '';
 336     my $cmd = "genlang -o -t=$target -e=$english $langfile 2>/dev/null";
 337     my $pool_file;
 338     open(VOICEFONTIDS, "> voicefontids");
 339     my $i = 0;
 340     local $| = 1; # make progress indicator work reliably
 341
 342     my $tts_object = init_tts($tts_engine, $tts_engine_opts, $language);
 343     print("Generating voice clips");
 344     print("\n") if $verbose;
 345     for (`$cmd`) {
 346         my $line = $_;
 347         print(VOICEFONTIDS $line);
 348         if ($line =~ /^id: (.*)$/) {
 349             $id = $1;
 350         }
 351         elsif ($line =~ /^voice: "(.*)"$/) {
 352             $voice = $1;
 353             if ($id !~ /^NOT_USED_.*$/ && $voice ne "") {
 354                 my $wav = $id . '.wav';
 355                 my $mp3 = $id . '.mp3';
 356
 357                 # Print some progress information
 358                 if (++$i % 10 == 0 and !$verbose) {
 359                     print(".");
 360                 }
 361
 362                 # Apply corrections to the string
 363                 $voice = correct_string($voice, $language, $tts_object);
 364
 365                 # If we have a pool of snippets, see if the string exists there first
 366                 if (defined($ENV{'POOL'})) {
 367                     $pool_file = sprintf("%s/%s-%s.mp3", $ENV{'POOL'},
 368                                          md5_hex("$voice $tts_engine $tts_engine_opts $encoder_opts"),
 369                                          $language);
 370                     if (-f $pool_file) {
 371                         printf("Re-using %s (%s) from pool\n", $id, $voice) if $verbose;
 372                         copy($pool_file, $mp3);
 373                     }
 374                 }
 375
 376                 # Don't generate MP3 if it already exists (probably from the POOL)
 377                 if (! -f $mp3) {
 378                     if ($id eq "VOICE_PAUSE") {
 379                         print("Use distributed $wav\n") if $verbose;
 380                         copy(dirname($0)."/VOICE_PAUSE.wav", $wav);
 381                     }
 382                     else {
 383                         voicestring($voice, $wav, $tts_engine_opts, $tts_object);
 384                         wavtrim($wav, 500, $tts_object);
 385                         # 500 seems to be a reasonable default for now
 386                     }
 387
 388                     encodewav($wav, $mp3, $encoder, $encoder_opts, $tts_object);
 389                     synchronize($tts_object);
 390                     if (defined($ENV{'POOL'})) {
 391                         copy($mp3, $pool_file);
 392                     }
 393                     unlink($wav);
 394                 }
 395                 $voice = "";
 396                 $id = "";
 397             }
 398         }
 399     }
 400     print("\n");
 401     close(VOICEFONTIDS);
 402     shutdown_tts($tts_object);
 403 }
 404
 405 # Assemble the voicefile
 406 sub createvoice {
 407     our $verbose;
 408     my ($language, $target_id) = @_;
 409     my $outfile = "";
 410     $outfile = sprintf("%s.voice", $language);
 411     printf("Saving voice file to %s\n", $outfile) if $verbose;
 412     my $cmd = "voicefont 'voicefontids' $target_id ./ $outfile";
 413     print("> $cmd\n") if $verbose;
 414     my $output = `$cmd`;
 415     print($output) if $verbose;
 416 }
 417
 418 sub deletemp3s() {
 419     for (glob('*.mp3')) {
 420         unlink($_);
 421     }
 422     for (glob('*.wav')) {
 423         unlink($_);
 424     }
 425 }
 426
 427 sub panic_cleanup {
 428     deletemp3s();
 429     die "moo";
 430 }
 431
 432 # Generate .talk clips
 433 sub gentalkclips {
 434     our $verbose;
 435     my ($dir, $tts_object, $encoder, $encoder_opts, $tts_engine_opts, $i) = @_;
 436     my $d = new DirHandle $dir;
 437     while (my $file = $d->read) {
 438         my ($voice, $wav, $mp3);
 439         # Print some progress information
 440         if (++$i % 10 == 0 and !$verbose) {
 441             print(".");
 442         }
 443
 444         # Convert to a complete path
 445         my $path = sprintf("%s/%s", $dir, $file);
 446
 447         $voice = $file;
 448         $wav = sprintf("%s.talk.wav", $path);
 449
 450         # Ignore dot-dirs and talk files
 451         if ($file eq '.' || $file eq '..' || $file =~ /\.talk$/) {
 452             next;
 453         }
 454         # Element is a dir
 455         if ( -d $path) {
 456             gentalkclips($path, $tts_object, $encoder, $encoder_opts, $tts_engine_opts, $i);
 457             $mp3 = sprintf("%s/_dirname.talk", $path);
 458         }
 459         # Element is a file
 460         else {
 461             $mp3 = sprintf("%s.talk", $path);
 462             $voice =~ s/\.[^\.]*$//; # Trim extension
 463         }
 464
 465         printf("Talkclip %s: %s", $mp3, $voice) if $verbose;
 466
 467         voicestring($voice, $wav, $tts_engine_opts, $tts_object);
 468         wavtrim($wav, 500, $tts_object);
 469         # 500 seems to be a reasonable default for now
 470         encodewav($wav, $mp3, $encoder, $encoder_opts, $tts_object);
 471         synchronize($tts_object);
 472         unlink($wav);
 473     }
 474 }
 475
 476
 477 # Check parameters
 478 my $printusage = 0;
 479 unless (defined($V) or defined($C)) { print("Missing either -V or -C\n"); $printusage = 1; }
 480 if (defined($V)) {
 481     unless (defined($t)) { print("Missing -t argument\n"); $printusage = 1; }
 482     unless (defined($l)) { print("Missing -l argument\n"); $printusage = 1; }
 483     unless (defined($i)) { print("Missing -i argument\n"); $printusage = 1; }
 484 }
 485 elsif (defined($C)) {
 486     unless (defined($ARGV[0])) { print "Missing path argument\n"; $printusage = 1; }
 487 }
 488 unless (defined($e)) { print("Missing -e argument\n"); $printusage = 1; }
 489 unless (defined($E)) { print("Missing -E argument\n"); $printusage = 1; }
 490 unless (defined($s)) { print("Missing -s argument\n"); $printusage = 1; }
 491 unless (defined($S)) { print("Missing -S argument\n"); $printusage = 1; }
 492 if ($printusage == 1) { printusage(); exit 1; }
 493
 494 if (defined($v) or defined($ENV{'V'})) {
 495     our $verbose = 1;
 496 }
 497
 498 # add the tools dir to the path temporarily, for calling various tools
 499 $ENV{'PATH'} = dirname($0) . ':' . $ENV{'PATH'};
 500
 501
 502 # Do what we're told
 503 if ($V == 1) {
 504     # Only do the panic cleanup for voicefiles
 505     $SIG{INT} = \&panic_cleanup;
 506     $SIG{KILL} = \&panic_cleanup;
 507
 508     printf("Generating voice\n  Target: %s\n  Language: %s\n  Encoder (options): %s (%s)\n  TTS Engine (options): %s (%s)\n",
 509         $t, $l, $e, $E, $s, $S);
 510     generateclips($l, $t, $e, $E, $s, $S);
 511     createvoice($l, $i);
 512     deletemp3s();
 513 }
 514 elsif ($C) {
 515     printf("Generating .talk clips\n  Path: %s\n  Language: %s\n  Encoder (options): %s (%s)\n  TTS Engine (options): %s (%s)\n", $ARGV[0], $l, $e, $E, $s, $S);
 516     my $tts_object = init_tts($s, $S, $l);
 517     gentalkclips($ARGV[0], $tts_object, $e, $E, $S, 0);
 518     shutdown_tts($tts_object);
 519 }
 520 else {
 521     printusage();
 522     exit 1;
 523 }