SynthesisEngineの抽象基底クラスSynthesisEngineBaseを追加 (#227)

* add SynthesisEngineBase

* use NotImplementedError

* add abstractmethod annotation

* add metaclass=ABCMeta

* add test_mock_synthesis_engine

* split test

* synthesis test

run.py

@@ -31,15 +31,15 @@
     synthesis_morphing_parameter as _synthesis_morphing_parameter,
 )
 from voicevox_engine.preset import Preset, PresetLoader
-from voicevox_engine.synthesis_engine import SynthesisEngine, make_synthesis_engine
+from voicevox_engine.synthesis_engine import SynthesisEngineBase, make_synthesis_engine
 from voicevox_engine.utility import ConnectBase64WavesException, connect_base64_waves
 
 
 def b64encode_str(s):
     return base64.b64encode(s).decode("utf-8")
 
 
-def generate_app(engine: SynthesisEngine) -> FastAPI:
+def generate_app(engine: SynthesisEngineBase) -> FastAPI:
     root_dir = Path(__file__).parent
 
     default_sampling_rate = engine.default_sampling_rate

test/test_mock_synthesis_engine.py

@@ -0,0 +1,140 @@
+from unittest import TestCase
+
+from voicevox_engine.dev.synthesis_engine import MockSynthesisEngine
+from voicevox_engine.kana_parser import create_kana
+from voicevox_engine.model import AccentPhrase, AudioQuery, Mora
+
+
+class TestMockSynthesisEngine(TestCase):
+    def setUp(self):
+        super().setUp()
+
+        self.accent_phrases_hello_hiho = [
+            AccentPhrase(
+                moras=[
+                    Mora(
+                        text="コ",
+                        consonant="k",
+                        consonant_length=0.0,
+                        vowel="o",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="ン",
+                        consonant=None,
+                        consonant_length=None,
+                        vowel="N",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="ニ",
+                        consonant="n",
+                        consonant_length=0.0,
+                        vowel="i",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="チ",
+                        consonant="ch",
+                        consonant_length=0.0,
+                        vowel="i",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="ワ",
+                        consonant="w",
+                        consonant_length=0.0,
+                        vowel="a",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                ],
+                accent=5,
+                pause_mora=Mora(
+                    text="、",
+                    consonant=None,
+                    consonant_length=None,
+                    vowel="pau",
+                    vowel_length=0.0,
+                    pitch=0.0,
+                ),
+            ),
+            AccentPhrase(
+                moras=[
+                    Mora(
+                        text="ヒ",
+                        consonant="h",
+                        consonant_length=0.0,
+                        vowel="i",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="ホ",
+                        consonant="h",
+                        consonant_length=0.0,
+                        vowel="o",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="デ",
+                        consonant="d",
+                        consonant_length=0.0,
+                        vowel="e",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                    Mora(
+                        text="ス",
+                        consonant="s",
+                        consonant_length=0.0,
+                        vowel="U",
+                        vowel_length=0.0,
+                        pitch=0.0,
+                    ),
+                ],
+                accent=1,
+                pause_mora=None,
+            ),
+        ]
+        self.engine = MockSynthesisEngine(speakers="")
+
+    def test_replace_phoneme_length(self):
+        self.assertEqual(
+            self.engine.replace_phoneme_length(
+                accent_phrases=self.accent_phrases_hello_hiho,
+                speaker_id=0,
+            ),
+            self.accent_phrases_hello_hiho,
+        )
+
+    def test_replace_mora_pitch(self):
+        self.assertEqual(
+            self.engine.replace_mora_pitch(
+                accent_phrases=self.accent_phrases_hello_hiho,
+                speaker_id=0,
+            ),
+            self.accent_phrases_hello_hiho,
+        )
+
+    def test_synthesis(self):
+        self.engine.synthesis(
+            AudioQuery(
+                accent_phrases=self.accent_phrases_hello_hiho,
+                speedScale=1,
+                pitchScale=0,
+                intonationScale=1,
+                volumeScale=1,
+                prePhonemeLength=0.1,
+                postPhonemeLength=0.1,
+                outputSamplingRate=24000,
+                outputStereo=False,
+                kana=create_kana(self.accent_phrases_hello_hiho),
+            ),
+            speaker_id=0,
+        )

test/test_mora_to_text.py

@@ -1,7 +1,7 @@
 from unittest import TestCase
 
 # TODO: import from voicevox_engine.synthesis_engine.mora
-from voicevox_engine.synthesis_engine.synthesis_engine import mora_to_text
+from voicevox_engine.synthesis_engine.synthesis_engine_base import mora_to_text
 
 
 class TestMoraToText(TestCase):

voicevox_engine/dev/synthesis_engine/__init__.py

@@ -1,3 +1,3 @@
-from .mock import SynthesisEngine
+from .mock import MockSynthesisEngine
 
-__all__ = ["SynthesisEngine"]
+__all__ = ["MockSynthesisEngine"]

voicevox_engine/dev/synthesis_engine/mock.py

@@ -6,10 +6,11 @@
 from scipy.signal import resample
 
 from ...model import AccentPhrase, AudioQuery
+from ...synthesis_engine import SynthesisEngineBase
 from ...synthesis_engine.synthesis_engine import to_flatten_moras
 
 
-class SynthesisEngine:
+class MockSynthesisEngine(SynthesisEngineBase):
     """
     SynthesisEngine [Mock]
     """

voicevox_engine/synthesis_engine/__init__.py

@@ -1,9 +1,11 @@
 from .forwarder import Forwarder
 from .make_synthesis_engine import make_synthesis_engine
 from .synthesis_engine import SynthesisEngine
+from .synthesis_engine_base import SynthesisEngineBase
 
 __all__ = [
     "Forwarder",
     "make_synthesis_engine",
     "SynthesisEngine",
+    "SynthesisEngineBase",
 ]

voicevox_engine/synthesis_engine/make_synthesis_engine.py

@@ -2,14 +2,14 @@
 from pathlib import Path
 from typing import Optional
 
-from .synthesis_engine import SynthesisEngine
+from .synthesis_engine import SynthesisEngine, SynthesisEngineBase
 
 
 def make_synthesis_engine(
     use_gpu: bool,
     voicelib_dir: Path,
     voicevox_dir: Optional[Path] = None,
-) -> SynthesisEngine:
+) -> SynthesisEngineBase:
     """
     音声ライブラリをロードして、音声合成エンジンを生成
 
@@ -57,7 +57,7 @@ def make_synthesis_engine(
             speakers=core.metas(),
         )
 
-    from ..dev.synthesis_engine import SynthesisEngine as MockSynthesisEngine
+    from ..dev.synthesis_engine import MockSynthesisEngine
 
     # モックで置き換える
     return MockSynthesisEngine(speakers=core.metas())

voicevox_engine/synthesis_engine/synthesis_engine.py

@@ -5,9 +5,8 @@
 from scipy.signal import resample
 
 from ..acoustic_feature_extractor import OjtPhoneme, SamplingData
-from ..full_context_label import extract_full_context_label
 from ..model import AccentPhrase, AudioQuery, Mora
-from ..mora_list import openjtalk_mora2text
+from .synthesis_engine_base import SynthesisEngineBase
 
 unvoiced_mora_phoneme_list = ["A", "I", "U", "E", "O", "cl", "pau"]
 mora_phoneme_list = ["a", "i", "u", "e", "o", "N"] + unvoiced_mora_phoneme_list
@@ -126,17 +125,7 @@ def pre_process(
     return flatten_moras, phoneme_data_list
 
 
-def mora_to_text(mora: str) -> str:
-    if mora[-1:] in ["A", "I", "U", "E", "O"]:
-        # 無声化母音を小文字に
-        mora = mora[:-1] + mora[-1].lower()
-    if mora in openjtalk_mora2text:
-        return openjtalk_mora2text[mora]
-    else:
-        return mora
-
-
-class SynthesisEngine:
+class SynthesisEngine(SynthesisEngineBase):
     def __init__(
         self,
         yukarin_s_forwarder,
@@ -365,72 +354,6 @@ def _create_one_hot(accent_phrase: AccentPhrase, position: int):
 
         return accent_phrases
 
-    def replace_mora_data(
-        self,
-        accent_phrases: List[AccentPhrase],
-        speaker_id: int,
-    ) -> List[AccentPhrase]:
-        return self.replace_mora_pitch(
-            accent_phrases=self.replace_phoneme_length(
-                accent_phrases=accent_phrases,
-                speaker_id=speaker_id,
-            ),
-            speaker_id=speaker_id,
-        )
-
-    def create_accent_phrases(self, text: str, speaker_id: int) -> List[AccentPhrase]:
-        if len(text.strip()) == 0:
-            return []
-
-        utterance = extract_full_context_label(text)
-        if len(utterance.breath_groups) == 0:
-            return []
-
-        return self.replace_mora_data(
-            accent_phrases=[
-                AccentPhrase(
-                    moras=[
-                        Mora(
-                            text=mora_to_text(
-                                "".join([p.phoneme for p in mora.phonemes])
-                            ),
-                            consonant=(
-                                mora.consonant.phoneme
-                                if mora.consonant is not None
-                                else None
-                            ),
-                            consonant_length=0 if mora.consonant is not None else None,
-                            vowel=mora.vowel.phoneme,
-                            vowel_length=0,
-                            pitch=0,
-                        )
-                        for mora in accent_phrase.moras
-                    ],
-                    accent=accent_phrase.accent,
-                    pause_mora=(
-                        Mora(
-                            text="、",
-                            consonant=None,
-                            consonant_length=None,
-                            vowel="pau",
-                            vowel_length=0,
-                            pitch=0,
-                        )
-                        if (
-                            i_accent_phrase == len(breath_group.accent_phrases) - 1
-                            and i_breath_group != len(utterance.breath_groups) - 1
-                        )
-                        else None
-                    ),
-                )
-                for i_breath_group, breath_group in enumerate(utterance.breath_groups)
-                for i_accent_phrase, accent_phrase in enumerate(
-                    breath_group.accent_phrases
-                )
-            ],
-            speaker_id=speaker_id,
-        )
-
     def synthesis(self, query: AudioQuery, speaker_id: int):
         """
         音声合成クエリから音声合成に必要な情報を構成し、実際に音声合成を行う