Use speaker names as well as ids

mimic3-tts/mimic3_tts/__main__.py

@@ -21,7 +21,23 @@ speaker = SSMLSpeaker(tts)
 # ssml = '<speak><voice name="uk_UK/m-ailabs_low"><s><w>бажав</w></s></voice></speak>'
 # ssml = '<speak><s><w>Hello</w><w>World</w></s></speak>'
 # ssml = '<speak><s>Hello world</s></speak>'
-ssml = '<speak><s><voice name="el_GR/rapunzelina_low"><say-as interpret-as="characters">12</say-as></voice></s></speak>'
+# ssml = '<speak><s><voice name="el_GR/rapunzelina_low"><say-as interpret-as="characters">12</say-as></voice></s></speak>'
+ssml = '''
+<speak>
+<voice name="en_US/amy_low">
+  <s>
+    Today is 1/2.
+    <break time="1s" />
+  </s>
+</voice>
+
+
+<voice name="es_ES/carlfm_low">
+  <s>Soy el <say-as interpret-as="number" format="ordinal">1</say-as>.</s>
+</voice>
+</speak>
+'''
+
 
 wav_file: wave.Wave_write = wave.open("out.wav", "wb")
 params_set = False

mimic3-tts/mimic3_tts/tts.py

@@ -27,7 +27,7 @@ from opentts_abc import (
 
 from mimic3_tts.config import TrainingConfig
 from mimic3_tts.utils import audio_float_to_int16
-from mimic3_tts.voice import Mimic3Voice
+from mimic3_tts.voice import Mimic3Voice, SPEAKER_TYPE
 
 _DIR = Path(__file__).parent
 
@@ -47,7 +47,7 @@ class Mimic3Settings:
     voice: typing.Optional[str] = None
     language: typing.Optional[str] = None
     voices_directories: typing.Optional[typing.Iterable[typing.Union[str, Path]]] = None
-    speaker_id: typing.Optional[int] = None
+    speaker: typing.Optional[SPEAKER_TYPE] = None
     length_scale: float = 1.0
     noise_scale: float = 0.667
     noise_w: float = 0.8
@@ -81,26 +81,24 @@ class Mimic3TextToSpeechSystem(TextToSpeechSystem):
     @voice.setter
     def voice(self, new_voice: str):
         if new_voice != self.settings.voice:
-            # Clear speaker id on voice change
+            # Clear speaker on voice change
-            self.speaker_id = None
+            self.speaker = None
 
         self.settings.voice = new_voice
 
         if "#" in self.settings.voice:
             # Split
-            voice, speaker_id_str = self.settings.voice.split("#", maxsplit=1)
+            voice, speaker = self.settings.voice.split("#", maxsplit=1)
             self.settings.voice = voice
-
+            self.speaker = speaker
-            # TODO: Use speaker map
-            self.speaker_id = int(speaker_id_str)
 
     @property
-    def speaker_id(self) -> typing.Optional[int]:
+    def speaker(self) -> typing.Optional[SPEAKER_TYPE]:
-        return self.settings.speaker_id
+        return self.settings.speaker
 
-    @speaker_id.setter
+    @speaker.setter
-    def speaker_id(self, new_speaker_id: typing.Optional[int]):
+    def speaker(self, new_speaker: typing.Optional[SPEAKER_TYPE]):
-        self.settings.speaker_id = new_speaker_id
+        self.settings.speaker = new_speaker
 
     @property
     def language(self) -> str:
@@ -154,12 +152,15 @@ class Mimic3TextToSpeechSystem(TextToSpeechSystem):
         for sent_phonemes in voice.text_to_phonemes(text, text_language=text_language):
             self._results.append(
                 Mimic3Phonemes(
-                    current_settings=deepcopy(self.settings), phonemes=sent_phonemes,
+                    current_settings=deepcopy(self.settings),
+                    phonemes=sent_phonemes,
                 )
             )
 
     def _speak_sentence_phonemes(
-        self, sent_phonemes, settings: typing.Optional[Mimic3Settings] = None,
+        self,
+        sent_phonemes,
+        settings: typing.Optional[Mimic3Settings] = None,
     ) -> AudioResult:
         settings = settings or self.settings
         voice = self._get_or_load_voice(settings.voice or self.voice)
@@ -169,7 +170,7 @@ class Mimic3TextToSpeechSystem(TextToSpeechSystem):
 
         audio = voice.ids_to_audio(
             sent_phoneme_ids,
-            speaker=self.speaker_id,
+            speaker=self.speaker,
             length_scale=settings.length_scale,
             noise_scale=settings.noise_scale,
             noise_w=settings.noise_w,

mimic3-tts/mimic3_tts/voice.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+import csv
 import itertools
 import logging
 import time
@@ -24,6 +25,7 @@ PHONEME_MAP_TYPE = typing.Dict[PHONEME_TYPE, typing.List[PHONEME_TYPE]]
 
 SPEAKER_NAME_TYPE = str
 SPEAKER_ID_TYPE = int
+SPEAKER_TYPE = typing.Union[SPEAKER_NAME_TYPE, SPEAKER_ID_TYPE]
 SPEAKER_MAP_TYPE = typing.Dict[SPEAKER_NAME_TYPE, SPEAKER_ID_TYPE]
 
 DEFAULT_LANGUAGE = "en_US"
@@ -137,7 +139,19 @@ class Mimic3Voice(metaclass=ABCMeta):
             speaker_id = 0
             if isinstance(speaker, SPEAKER_NAME_TYPE):
                 if self.speaker_map:
-                    speaker_id = self.speaker_map.get(speaker, speaker_id)
+                    maybe_speaker_id = self.speaker_map.get(speaker)
+                    if maybe_speaker_id is None:
+                        try:
+                            # Interpret as speaker id
+                            speaker_id = int(speaker)
+                        except ValueError:
+                            _LOGGER.warning(
+                                "Unable to find a speaker with the name '%s'. Falling back to first speaker.",
+                                speaker,
+                            )
+                            pass
+                    else:
+                        speaker_id = maybe_speaker_id
             elif speaker is not None:
                 speaker_id = speaker
 
@@ -198,7 +212,18 @@ class Mimic3Voice(metaclass=ABCMeta):
             with open(phoneme_map_path, "r", encoding="utf-8") as map_file:
                 phoneme_map = phonemes2ids.utils.load_phoneme_map(map_file)
 
-        # TODO: Load speaker map
+        # id -> speaker | alias | alias ...
+        speaker_map: typing.Optional[SPEAKER_MAP_TYPE] = None
+        speaker_map_path = voice_dir / "speaker_map.csv"
+        if speaker_map_path.is_file():
+            _LOGGER.debug("Loading speaker map from %s", speaker_map_path)
+            with open(speaker_map_path, "r", encoding="utf-8") as map_file:
+                reader = csv.reader(map_file, delimiter="|")
+                speaker_map = {}
+                for row in reader:
+                    speaker_id = int(row[0])
+                    for alias in row[1:]:
+                        speaker_map[alias] = speaker_id
 
         if config.phonemizer == Phonemizer.GRUUT:
             return GruutVoice(
@@ -206,6 +231,7 @@ class Mimic3Voice(metaclass=ABCMeta):
                 onnx_model=onnx_model,
                 phoneme_to_id=phoneme_to_id,
                 phoneme_map=phoneme_map,
+                speaker_map=speaker_map,
             )
 
         if config.phonemizer == Phonemizer.ESPEAK:
@@ -214,6 +240,7 @@ class Mimic3Voice(metaclass=ABCMeta):
                 onnx_model=onnx_model,
                 phoneme_to_id=phoneme_to_id,
                 phoneme_map=phoneme_map,
+                speaker_map=speaker_map,
             )
         if config.phonemizer == Phonemizer.SYMBOLS:
             return SymbolsVoice(
@@ -221,6 +248,7 @@ class Mimic3Voice(metaclass=ABCMeta):
                 onnx_model=onnx_model,
                 phoneme_to_id=phoneme_to_id,
                 phoneme_map=phoneme_map,
+                speaker_map=speaker_map,
             )
 
         raise ValueError(f"Unsupported phonemizer: {config.phonemizer}")