First pass at Dutch IPA

app.py

@@ -363,7 +363,7 @@ def api_pronounce() -> Union[Response, str]:
 
     if pronounce_type == "phonemes":
         # Convert from Sphinx to espeak phonemes
-        espeak_str = core.get_word_phonemes(pronounce_str).phonemes
+        espeak_str = core.get_word_phonemes(pronounce_str).phonemes["espeak"]
     else:
         # Speak word directly
         espeak_str = pronounce_str

profiles/defaults.json

@@ -159,7 +159,8 @@
             "program": ""
         },
         "espeak": {
-            "phoneme_map": "espeak_phonemes.txt"
+            "phoneme_map": "espeak_phonemes.txt",
+            "ipa_map": "ipa_phonemes.txt"
         },
         "flite": {
             "voice": "kal16"

profiles/es/ipa_examples.txt

@@ -0,0 +1,25 @@
+a abre  əbɹˈɛ
+b baja  bˈɑhɑ
+ch ocho  ˈɒtʃəʊ
+d anda  ˈændə
+e aire  ˈɛə
+f café  kæfˈeɪ
+g agua  ˈæɡjʉːə
+gn años  sˈænsəːn
+i allí  ˈælɪ
+j bajo  bˈɑdʒəʊ
+k acto  ˈæktəʊ
+l alex  ˈælɪks
+ll allá  əlˈɑ
+m alma  ˈɒlmə
+n andy  ˈændɪ
+o algo  ˈælɡəʊ
+p copa  kˈəʊpə
+r amor  ˈæmɚ
+rr raro  ɹˈæɹəʊ
+s base  bˈeɪs
+t alta  ˈɔːltə
+u auto  ˈɔːtəʊ
+x sexo  sˈiksəʊ
+y ayer  ˈaɪɚ
+z azul  æzˈʉːl

profiles/es/ipa_phonemes.txt

@@ -0,0 +1,25 @@
+a abre  a
+b baja  b
+ch ocho  tʃ
+d anda  d
+e aire  e
+f café  f
+g agua  g
+gn años ɲ
+i allí  i
+j bajo  j
+k acto  k
+l alex  l
+ll allá  ʎ
+m alma  m
+n andy  n
+o algo  o
+p copa  p
+r amor  r
+rr raro ɾ
+s base  s
+t alta  t
+u auto  u
+x sexo  x
+y ayer  ʝ
+z azul  z

profiles/es/phoneme_examples.txt

@@ -1,22 +1,25 @@
-A A A
-B VEZ B E Z
-CH MUCHO M U CH O
-D DE D E
-E EL E L
-F FUE F U E
-G ALGO A L G O
-GN AñOS A GN O S
-I Y I
-K CON K O N
-L AL A L
-LL ELLA E LL A
-M ME M E
-N EN E N
-O O O
-P POR P O R
-R ERA E R A
-S ES E S
-T TE T E
-U SU S U
-Y YA Y A
-Z HACE A Z E
+a abre a b r e
+b baja b a j a
+ch ocho o ch o
+d anda a n d a
+e aire a i r e
+f café k a f e
+g agua a g u a
+gn años a gn o s
+i allí a ll i
+j bajo b a j o
+k acto a k t o
+l alex a l e x
+ll allá a ll a
+m alma a l m a
+n andy a n d i
+o algo a l g o
+p copa k o p a
+r amor a m o r
+rr raro rr a r o
+s base b a s e
+t alta a l t a
+u auto a u t o
+x sexo s e x o
+y ayer a y e r
+z azul a z u l

profiles/nl/ipa_examples.txt

@@ -0,0 +1,38 @@
+@ adem  ə
+a adam  aː
+aa acht  ɑ
+b baan  b
+d andy  d
+e baby  eː
+ee alex  ɛ
+ei blij  ɛi
+er albert  ɛː
+eu deur  øː
+f cafÉ  f
+g chicago  g
+gg hoge  ɟ
+h haag  ɦ
+i bier  i
+ii bill  ɪ
+j doei  j
+k bank  k
+l alle  l
+m amen  m
+n anna  n
+nn bang  ŋ
+o auto  oː
+oo bond  ɔ
+ou bouw  ʌu
+p club  p
+r arme  r
+s arts  s
+ss chef  ʃ
+t band  t
+u boek  u
+ui buik  œy
+v dave  v
+w eeuw  ʋ
+x berg  x
+y duur  y
+yy brug  ʒ
+z deze  z

rhasspy/__init__.py

@@ -1035,7 +1035,8 @@ def word2wav(core: RhasspyCore, profile: Profile, args: Any) -> None:
     word_pronunciations = all_pronunciations[args.word]["pronunciations"]
 
     # Convert from CMU phonemes to eSpeak phonemes
-    espeak_str = core.get_word_phonemes(word_pronunciations[0]).phonemes
+    sphinx = word_pronunciations[0]
+    espeak_str = all_pronunciations[args.word]["phonemes"][sphinx]["espeak"]
 
     # Pronounce as WAV
     wav_data = core.speak_word(espeak_str).wav_data

rhasspy/pronounce.py

@@ -5,6 +5,7 @@ import logging
 import subprocess
 import tempfile
 from typing import Dict, Tuple, List, Optional, Any
+from collections import defaultdict
 
 from .actor import RhasspyActor
 from .utils import read_dict, load_phoneme_map
@@ -35,7 +36,7 @@ class GetWordPhonemes:
 
 
 class WordPhonemes:
-    def __init__(self, word: str, phonemes: str) -> None:
+    def __init__(self, word: str, phonemes: Dict[str, str]) -> None:
         self.word = word
         self.phonemes = phonemes
 
@@ -98,8 +99,11 @@ class PhonetisaurusPronounce(RhasspyActor):
                 self._logger.exception("pronounce")
                 self.send(message.receiver or sender, PronunciationFailed(repr(e)))
         elif isinstance(message, GetWordPhonemes):
-            phonemes = self.translate_phonemes(message.word)
-            self.send(message.receiver or sender, WordPhonemes(message.word, phonemes))
+            phonemes = self.translate_phonemes([message.word])
+            self.send(
+                message.receiver or sender,
+                WordPhonemes(message.word, phonemes[message.word]),
+            )
 
     # -------------------------------------------------------------------------
 
@@ -129,20 +133,49 @@ class PhonetisaurusPronounce(RhasspyActor):
 
     # -------------------------------------------------------------------------
 
-    def translate_phonemes(self, phonemes: str) -> str:
-        # Load map from Sphinx to eSpeak phonemes
+    def translate_phonemes(self, sphinxes: List[str]) -> Dict[str, Dict[str, str]]:
+        from .lexconvert import convert
+
+        # Load map from Sphinx to IPA
         map_path = self.profile.read_path(
-            self.profile.get("text_to_speech.espeak.phoneme_map")
+            self.profile.get(
+                "text_to_speech.espeak.ipa.phoneme_map", "ipa_phonemes.txt"
+            )
         )
 
-        phoneme_map = load_phoneme_map(map_path)
+        if os.path.exists(map_path):
+            # Compute Sphinx <-> eSpeak map using IPA
+            ipa_map = load_phoneme_map(map_path)
+            phoneme_map = {
+                sphinx: convert(ipa, "unicode-ipa", "espeak")
+                for sphinx, ipa in ipa_map.items()
+            }
+        else:
+            # Fall back to Sphinx <-> eSpeak map
+            map_path = self.profile.read_path(
+                self.profile.get(
+                    "text_to_speech.espeak.phoneme_map", "espeak_phonemes.txt"
+                )
+            )
 
-        # Convert from Sphinx to espeak phonemes
-        espeak_str = "[['%s]]" % "".join(
-            phoneme_map.get(p, p) for p in phonemes.split()
-        )
+            phoneme_map = load_phoneme_map(map_path)
+            ipa_map = {
+                sphinx: convert(espeak, "espeak", "unicode-ipa")
+                for sphinx, espeak in phoneme_map.items()
+            }
 
-        return espeak_str
+        phonemes = defaultdict(dict)
+        for sphinx in sphinxes:
+            sphinx_parts = sphinx.split()
+
+            # Convert from Sphinx to espeak phonemes
+            phonemes[sphinx]["espeak"] = "[['%s]]" % "".join(
+                phoneme_map.get(p, p) for p in sphinx_parts
+            )
+
+            phonemes[sphinx]["ipa"] = "".join(ipa_map.get(p, p) for p in sphinx_parts)
+
+        return phonemes
 
     # -------------------------------------------------------------------------
 
@@ -168,20 +201,16 @@ class PhonetisaurusPronounce(RhasspyActor):
                     read_dict(dictionary_file, word_dict)
 
         pronunciations = self._lookup_words(words, word_dict, n)
+        all_pronunciations = []
+        for pron in pronunciations.values():
+            all_pronunciations.extend(pron["pronunciations"])
 
-        # Get phonemes from eSpeak
-        for word in words:
-            espeak_command = ["espeak", "-q", "-x"]
+        phonemes = self.translate_phonemes(all_pronunciations)
 
-            voice = self._get_voice()
-            if voice is not None:
-                espeak_command.extend(["-v", voice])
-
-            espeak_command.append(word)
-
-            self._logger.debug(repr(espeak_command))
-            espeak_str = subprocess.check_output(espeak_command).decode().strip()
-            pronunciations[word]["phonemes"] = espeak_str
+        for word in pronunciations.keys():
+            pronunciations[word]["phonemes"] = {}
+            for pron in pronunciations[word]["pronunciations"]:
+                pronunciations[word]["phonemes"][pron] = phonemes[pron]
 
         return pronunciations
 

src/components/LookupPronounce.vue

@@ -40,7 +40,7 @@
                     <button type="button" class="btn btn-success" title="Add this pronunciation to your custom words" @click="addToCustomWords">Add</button>
                 </div>
                 <div class="col-xs-auto">
-                    <input id="espeak-phonemes" title="eSpeak Phonemes" class="form-control" type="text" v-model="espeakPhonemes" readonly>
+                    <input id="espeak-phonemes" title="IPA/eSpeak Phonemes" class="form-control" type="text" v-model="espeakPhonemes" readonly>
                 </div>
                 <div class="col-xs-auto">
                     <button type="button" class="btn btn-secondary" @click="pronouncePhonemesOrWord(phonemes)"
@@ -176,7 +176,7 @@
                          this.phonemes = this.pronunciations[0];
                      }
 
-                     this.espeakPhonemes = request.data.phonemes
+                     this.espeakPhonemes = request.data.phonemes[this.phonemes]['ipa']
                  })
                  .then(() => this.$parent.endAsync())
                  .catch(err => this.$parent.error(err))