voice: Add support for the Piper TTS engine

   https://github.com/rhasspy/piper

High quality, offline, neural-network-based, with good language coverage

Note that you have to manually download the piper voice models, and set
PIPER_MODEL_DIR appropriately.  The configure script will let you choose
from the available models and remember your choices.

Change-Id: I8eba9fcf78b51b01b89491539aac3e423cc42f16

2 files changed, 147 insertions, 51 deletions

diff --git a/tools/configure b/tools/configure
index adccc3e5e4..51a47971fd 100755
--- a/tools/configure
+++ b/tools/configure
@@ -1159,6 +1159,13 @@ voiceconfig () {
         DEFAULT_TTS_OPTS=$GTTS_OPTS
         DEFAULT_CHOICE="g"
     fi
+    if [ -n "`findtool piper`" ]; then
+        PIPER="(p)iper "
+        PIPER_OPTS=""
+        DEFAULT_TTS="piper"
+        DEFAULT_TTS_OPTS=$PIPER_OPTS
+        DEFAULT_CHOICE="p"
+    fi
     if [ -n "`findtool rbspeak`" ]; then
         RBSPEAK="(O)ther "
         RBSPEAK_OPTS=""
@@ -1167,15 +1174,15 @@ voiceconfig () {
         DEFAULT_CHOICE="O"
     fi
 
-    if [ "$FESTIVAL" = "$FLITE" ] && [ "$FLITE" = "$ESPEAK" ] && [ "$ESPEAK" = "$SAPI" ] && [ "$SAPI" = "$MIMIC" ] && [ "$MIMIC" = "$SWIFT" ] && [ "$SWIFT" = "$GTTS" ] && [ "$GTTS" = "$RBSPEAK" ] ; then
-        echo "You need Festival, eSpeak, Mimic, Flite, gtts, or rbspeak in your path, or SAPI available to build voice files"
+    if [ "$FESTIVAL" = "$FLITE" ] && [ "$FLITE" = "$ESPEAK" ] && [ "$ESPEAK" = "$SAPI" ] && [ "$SAPI" = "$MIMIC" ] && [ "$MIMIC" = "$SWIFT" ] && [ "$SWIFT" = "$GTTS" ] && [ "$GTTS" = "$PIPER" ] && [ "$PIPER" = "$RBSPEAK" ] ; then
+        echo "You need Festival, eSpeak, Mimic, Flite, piper, gtts, or rbspeak in your path, or SAPI available to build voice files"
         exit 3
     fi
 
     if [ "$ARG_TTS" ]; then
         option=$ARG_TTS
     else
-        echo "TTS engine to use: ${FLITE}${FESTIVAL}${ESPEAK}${MIMIC}${SAPI}${SWIFT}${GTTS}${RBSPEAK}(${DEFAULT_CHOICE})?"
+        echo "TTS engine to use: ${FLITE}${FESTIVAL}${ESPEAK}${MIMIC}${SAPI}${SWIFT}${GTTS}${RBSPEAK}${PIPER}(${DEFAULT_CHOICE})?"
         option=`input`
         if [ -z "$option" ]; then option=${DEFAULT_CHOICE}; fi
         advopts="$advopts --tts=$option"
@@ -1209,6 +1216,10 @@ voiceconfig () {
         TTS_ENGINE="gtts"
         TTS_OPTS=$GTTS_OPTS
         ;;
+        [Pp]|piper)
+        TTS_ENGINE="piper"
+        TTS_OPTS=$PIPER_OPTS
+        ;;
         [Oo]|rbspeak)
         TTS_ENGINE="rbspeak"
         TTS_OPTS=$RBSPEAK_OPTS
@@ -1247,6 +1258,39 @@ voiceconfig () {
         advopts="$advopts --voice=$CHOICE"
         echo "Festival voice set to $TTS_FESTIVAL_VOICE"
         echo "(voice_$TTS_FESTIVAL_VOICE)" > festival-prolog.scm
+    elif [ "$TTS_ENGINE" = "piper" ]; then
+        if [ -z "$PIPER_MODEL_DIR" ]; then
+            echo "Please set PIPER_MODEL_DIR!";
+            exit 1
+        fi
+        models=`(cd $PIPER_MODEL_DIR ; ls -1 *onnx)`
+        for model in $models; do
+            PIPER_MODEL="$model"  # Default
+            break;
+        done
+        if [ "$ARG_VOICE" ]; then
+            CHOICE=$ARG_VOICE
+        else
+            i=1
+            for model in $models; do
+                printf "%3d. %s\n" "$i" "$model"
+                i=`expr $i + 1`
+            done
+            printf "Please select which piper model to use (default is $PIPER_MODEL): "
+            CHOICE=`input`
+        fi
+        i=1
+        for model in $models; do
+            if [ "$i" = "$CHOICE" -o "$model" = "$CHOICE" ]; then
+                PIPER_MODEL="$model"
+                break;
+            fi
+            i=`expr $i + 1`
+        done
+
+        TTS_OPTS="$TTS_OPTS --model $PIPER_MODEL_DIR/$PIPER_MODEL"
+        advopts="$advopts --voice=$PIPER_MODEL"
+        echo "Piper model set to $PIPER_MODEL"
     elif [ "$TTS_ENGINE" = "mimic" ]; then
         voicelist=`mimic -lv | cut -d':' -f2`
         for voice in $voicelist; do
@@ -1268,6 +1312,7 @@ voiceconfig () {
         for voice in $voicelist; do
             if [ "$i" = "$CHOICE" -o "$voice" = "$CHOICE" ]; then
                 TTS_MIMIC_VOICE="$voice"
+                break
             fi
             i=`expr $i + 1`
         done
@@ -4756,6 +4801,7 @@ export ANDROID_NDK_PATH=${ANDROID_NDK_PATH}
 export ANDROID_SDK_PATH=${ANDROID_SDK_PATH}
 export ANDROID_PLATFORM_VERSION=${ANDROID_PLATFORM_VERSION}
 export TOOLSET=${toolset}
+export PIPER_MODEL_DIR=${PIPER_MODEL_DIR}
 $CCACHE_ARG
 
 CONFIGURE_OPTIONS=${cmdline}
diff --git a/tools/voice.pl b/tools/voice.pl
index 6b3f807854..337407e187 100755
--- a/tools/voice.pl
+++ b/tools/voice.pl
@@ -42,7 +42,8 @@ Usage: voice.pl [options] [path to dir]
     Specify which target you want to build voicefile for. Must include
     any features that target supports.
 
- -f=<file> Use existing voiceids file
+ -f=<file>
+    Use existing voiceids file
 
  -i=<target_id>
     Numeric target id. Needed for voice building.
@@ -64,7 +65,8 @@ Usage: voice.pl [options] [path to dir]
     Options to pass to the TTS engine. Enclose in double quotes if the
     options include spaces.
 
- -F Force the file to be regenerated even if present
+ -F
+    Force the file to be regenerated even if present
 
  -v
     Be verbose
@@ -73,57 +75,78 @@ USAGE
 }
 
 my %festival_lang_map = (
-                           'english' => 'english',
-                           'english-us' => 'english',
-                           'espanol' => 'spanish',
-                          #'finnish' => 'finnish'
-                          #'italiano' => 'italian',
-                          #'czech' => 'czech',
-                          #'welsh' => 'welsh'
+    'english' => 'english',
+    'english-us' => 'english',
+    'espanol' => 'spanish',
+    #'finnish' => 'finnish'
+    #'italiano' => 'italian',
+    #'czech' => 'czech',
+    #'welsh' => 'welsh'
 );
 
 my %gtts_lang_map = (
     'english' => '-l en -t co.uk',  # Always first, it's the golden master
-        'czech' => '-l cs',   # not supported
-        'dansk' => '-l da',
-        'deutsch' => '-l de',
-        'english-us' => '-l en -t us',
-        'espanol' => '-l es',
-        'francais' => '-l fr',
-        'greek' => '-l el',
-        'magyar' => '-l hu',
-        'italiano' => '-l it',
-        'nederlands' => '-l nl',
-        'norsk' => '-l no',
-        'polski' => '-l pl',
-        'russian' => '-l ru',
-        'slovak' => '-l sk',
-        'srpski' => '-l sr',
-        'svenska' => '-l sv',
-        'turkce' => '-l tr',
+    'czech' => '-l cs',
+    'dansk' => '-l da',
+    'deutsch' => '-l de',
+    'english-us' => '-l en -t us',
+    'espanol' => '-l es',
+    'francais' => '-l fr',
+    'greek' => '-l el',
+    'magyar' => '-l hu',
+    'italiano' => '-l it',
+    'nederlands' => '-l nl',
+    'norsk' => '-l no',
+    'polski' => '-l pl',
+    'russian' => '-l ru',
+    'slovak' => '-l sk',
+    'srpski' => '-l sr',
+    'svenska' => '-l sv',
+    'turkce' => '-l tr',
 );
 
 my %espeak_lang_map = (
-    'english' => 'en-gb',  # Always first, it's the golden master
-        'czech' => 'cs',
-        'dansk' => 'da',
-        'deutsch' => 'de',
-        'english-us' => 'en-us',
-        'espanol' => 'es',
-        'francais' => 'fr-fr',
-        'greek' => 'el',
-        'nederlands' => 'nl',
-        'magyar' => 'hu',
-        'italiano' => 'it',
-        'japanese' => 'ja',
-        'nederlands' => 'nl',
-        'norsk' => 'no',
-        'polski' => 'pl',
-        'russian' => 'ru',
-        'slovak' => 'sk',
-        'srpski' => 'sr',
-        'svenska' => 'sv',
-        'turkce' => 'tr',
+    'english' => '-ven-gb -k 5',  # Always first, it's the golden master
+    'czech' => '-vcs',
+    'dansk' => '-vda',
+    'deutsch' => '-vde',
+    'english-us' => '-ven-us -k 5',
+    'espanol' => '-ves',
+    'francais' => '-vfr-fr',
+    'greek' => '-vel',
+    'magyar' => '-vhu',
+    'italiano' => '-vit',
+    'japanese' => '-vja',
+    'nederlands' => '-vnl',
+    'norsk' => '-vno',
+    'polski' => '-vpl',
+    'russian' => '-vru',
+    'slovak' => '-vsk',
+    'srpski' => '-vsr',
+    'svenska' => '-vsv',
+    'turkce' => '-vtr',
+    );
+
+my %piper_lang_map = (
+    'english' => 'en_GB-cori-high.onnx',  # Always first, it's the golden master
+    'czech' => 'cs_CZ-jirka-medium.onnx',
+    'dansk' => 'da_DK-talesyntese-medium.onnx',
+    'deutsch' => 'de_DE-thorsten-high.onnx',
+    'english-us' => 'en_US-libritts-high.onnx',
+    'espanol' => 'es_ES-sharvard-medium.onnx',
+    'francais' => 'fr_FR-siwis-medium.onnx',
+    'greek' => 'el_GR-rapunzelina-low.onnx',
+#    'magyar' => '-vhu',
+    'italiano' => 'it_IT-riccardo-x_low.onnx',
+#    'japanese' => '-vja',
+    'nederlands' => 'nl_NL-mls-medium.onnx',
+    'norsk' => 'no_NO-talesyntese-medium.onnx',
+    'polski' => 'pl_PL-gosia-medium.onnx',
+    'russian' => 'ru_RU-irina-medium.onnx',
+    'slovak' => 'sk_SK-lili-medium.onnx',
+    'srpski' => 'sr_RS-serbski_institut-medium.onnx',
+    'svenska' => 'sv_SE-nst-medium.onnx',
+    'turkce' => 'tr_TR-fettah-medium.onnx',
 );
 
 my $trim_thresh = 500;   # Trim silence if over this, in ms
@@ -141,6 +164,7 @@ sub init_tts {
     # Don't use given/when here - it's not compatible with old perl versions
     if ($tts_engine eq 'festival') {
         print("> festival $tts_engine_opts --server\n") if $verbose;
+        # Open command, and filehandles for STDIN, STDOUT, STDERR
         my $pid = open(FESTIVAL_SERVER, "| festival $tts_engine_opts --server > /dev/null 2>&1");
         my $dummy = *FESTIVAL_SERVER; #suppress warning
         $SIG{INT} = sub { kill TERM => $pid; print("foo"); panic_cleanup(); };
@@ -149,6 +173,21 @@ sub init_tts {
         if (defined($festival_lang_map{$language}) && $tts_engine_opts !~ /--language/) {
             $ret{"ttsoptions"} = "--language $festival_lang_map{$language} ";
         }
+    } elsif ($tts_engine eq 'piper') {
+        my $cmd = "piper $tts_engine_opts --json-input";
+        print("> $cmd\n") if $verbose;
+
+        my $pid = open3(*CMD_IN, *CMD_OUT, *CMD_ERR, $cmd);
+        $SIG{INT} = sub { kill TERM => $pid; print("foo"); panic_cleanup(); };
+        $SIG{KILL} = sub { kill TERM => $pid; print("boo"); panic_cleanup(); };
+        $ret{"pid"} = $pid;
+        binmode(*CMD_IN, ':encoding(utf8)');
+        binmode(*CMD_OUT, ':encoding(utf8)');
+        binmode(*CMD_ERR, ':encoding(utf8)');
+        if (defined($piper_lang_map{$language}) && $tts_engine_opts !~ /--model/) {
+            die("Need PIPER_MODEL_DIR\n") if (!defined($ENV{'PIPER_MODEL_DIR'}));
+            $ret{"ttsoptions"} = "--model $ENV{PIPER_MODEL_DIR}/$piper_lang_map{$language} ";
+        }
     } elsif ($tts_engine eq 'sapi') {
         my $toolsdir = dirname($0);
         my $path = `cygpath $toolsdir -a -w`;
@@ -176,7 +215,7 @@ sub init_tts {
         }
     } elsif ($tts_engine eq 'espeak' || $tts_engine eq 'espeak-ng') {
         if (defined($espeak_lang_map{$language}) && $tts_engine_opts !~ /-v/) {
-            $ret{"ttsoptions"} = "-v$espeak_lang_map{$language} ";
+            $ret{"ttsoptions"} = " $espeak_lang_map{$language} ";
         }
     }
 
@@ -190,6 +229,10 @@ sub shutdown_tts {
         # Send SIGTERM to festival server
         kill TERM => $$tts_object{"pid"};
     }
+    elsif ($$tts_object{'name'} eq 'piper') {
+        # Send SIGTERM to piper
+        kill TERM => $$tts_object{"pid"};
+    }
     elsif ($$tts_object{'name'} eq 'sapi') {
         print({$$tts_object{"stdin"}} "QUIT\r\n");
         close($$tts_object{"stdin"});
@@ -244,6 +287,13 @@ sub voicestring {
         close(CMD_OUT);
         close(CMD_ERR);
     }
+    elsif ($name eq 'piper') {
+        $cmd = "{ \"text\": \"$string\", \"output_file\": \"$output\" }";
+        print(">> $cmd\n") if $verbose;
+        print(CMD_IN "$cmd\n");
+        my $res = <CMD_OUT>;
+        $res = <CMD_ERR>;
+    }
     elsif ($name eq 'flite') {
         $cmd = "flite $tts_engine_opts -t \"$string\" \"$output\"";
         print("> $cmd\n") if $verbose;
@@ -469,7 +519,6 @@ sub generateclips {
     print("\n");
 
     unlink($updfile) if (-f $updfile);
-    shutdown_tts($tts_object);
 }
 
 # Assemble the voicefile
@@ -608,6 +657,7 @@ if ($V == 1) {
            defined($t) ? $t : "unknown",
            $l, $e, $E, $s, $S);
     generateclips($l, $t, $e, $E, $tts_object, $S, $f);
+    shutdown_tts($tts_object);
     createvoice($l, $i, $f);
     deleteencs();
 } elsif ($C) {