refactor: tts_engine から song_engine に分離 (#1592)

Co-authored-by: Hiroshiba <[email protected]>

Author: raa0121

run.py

@@ -21,6 +21,7 @@
 from voicevox_engine.preset.preset_manager import PresetManager
 from voicevox_engine.setting.model import CorsPolicyMode
 from voicevox_engine.setting.setting_manager import USER_SETTING_PATH, SettingHandler
+from voicevox_engine.tts_pipeline.song_engine import make_song_engines_from_cores
 from voicevox_engine.tts_pipeline.tts_engine import make_tts_engines_from_cores
 from voicevox_engine.user_dict.user_dict_manager import UserDictionary
 from voicevox_engine.utility.path_utility import (
@@ -326,7 +327,9 @@ def main() -> None:
         load_all_models=args.load_all_models,
     )
     tts_engines = make_tts_engines_from_cores(core_manager)
+    song_engines = make_song_engines_from_cores(core_manager)
     assert len(tts_engines.versions()) != 0, "音声合成エンジンがありません。"
+    assert len(song_engines.versions()) != 0, "音声合成エンジンがありません。"
 
     cancellable_engine: CancellableEngine | None = None
     if args.enable_cancellable_synthesis:
@@ -389,6 +392,7 @@ def main() -> None:
     # ASGI に準拠した VOICEVOX ENGINE アプリケーションを生成する
     app = generate_app(
         tts_engines,
+        song_engines,
         core_manager,
         setting_loader,
         preset_manager,

test/benchmark/engine_preparation.py

@@ -13,6 +13,7 @@
 from voicevox_engine.library.library_manager import LibraryManager
 from voicevox_engine.preset.preset_manager import PresetManager
 from voicevox_engine.setting.setting_manager import SettingHandler
+from voicevox_engine.tts_pipeline.song_engine import make_song_engines_from_cores
 from voicevox_engine.tts_pipeline.tts_engine import make_tts_engines_from_cores
 from voicevox_engine.user_dict.user_dict_manager import UserDictionary
 from voicevox_engine.utility.path_utility import engine_manifest_path, get_save_dir
@@ -23,6 +24,7 @@ def _generate_engine_fake_server(root_dir: Path) -> TestClient:
         voicevox_dir=root_dir, use_gpu=False, enable_mock=False
     )
     tts_engines = make_tts_engines_from_cores(core_manager)
+    song_engines = make_song_engines_from_cores(core_manager)
     setting_loader = SettingHandler(Path("./not_exist.yaml"))
     preset_manager = PresetManager(get_save_dir() / "presets.yaml")
     user_dict = UserDictionary()
@@ -36,6 +38,7 @@ def _generate_engine_fake_server(root_dir: Path) -> TestClient:
     )
     app = generate_app(
         tts_engines=tts_engines,
+        song_engines=song_engines,
         core_manager=core_manager,
         setting_loader=setting_loader,
         preset_manager=preset_manager,

test/e2e/conftest.py

@@ -14,6 +14,7 @@
 from voicevox_engine.library.library_manager import LibraryManager
 from voicevox_engine.preset.preset_manager import PresetManager
 from voicevox_engine.setting.setting_manager import SettingHandler
+from voicevox_engine.tts_pipeline.song_engine import make_song_engines_from_cores
 from voicevox_engine.tts_pipeline.tts_engine import make_tts_engines_from_cores
 from voicevox_engine.user_dict.user_dict_manager import (
     DEFAULT_DICT_PATH,
@@ -33,6 +34,7 @@ def _copy_under_dir(file_path: Path, dir_path: Path) -> Path:
 def app_params(tmp_path: Path) -> dict[str, Any]:
     core_manager = initialize_cores(use_gpu=False, enable_mock=True)
     tts_engines = make_tts_engines_from_cores(core_manager)
+    song_engines = make_song_engines_from_cores(core_manager)
     setting_loader = SettingHandler(tmp_path / "not_exist.yaml")
 
     # テスト用に隔離されたプリセットを生成する
@@ -57,6 +59,7 @@ def app_params(tmp_path: Path) -> dict[str, Any]:
 
     return {
         "tts_engines": tts_engines,
+        "song_engines": song_engines,
         "core_manager": core_manager,
         "setting_loader": setting_loader,
         "preset_manager": preset_manager,

test/unit/tts_pipeline/test_tts_engine.py

@@ -17,6 +17,9 @@
     Note,
     Score,
 )
+from voicevox_engine.tts_pipeline.song_engine import (
+    SongEngine,
+)
 from voicevox_engine.tts_pipeline.tts_engine import (
     TTSEngine,
     _apply_interrogative_upspeak,
@@ -277,13 +280,13 @@ def test_mocked_create_sing_volume_from_phoneme_and_f0_output(
     NOTE: 入力生成の簡略化に別関数を呼び出すため、別関数が正しく動作しない場合テストが落ちる
     """
     # Inputs
-    tts_engine = TTSEngine(MockCoreWrapper())
+    tts_engine = SongEngine(MockCoreWrapper())
     doremi_srore = _gen_doremi_score()
-    phonemes, f0s, _ = tts_engine.create_sing_phoneme_and_f0_and_volume(
+    phonemes, f0s, _ = tts_engine.create_phoneme_and_f0_and_volume(
         doremi_srore, StyleId(1)
     )
     # Outputs
-    result = tts_engine.create_sing_volume_from_phoneme_and_f0(
+    result = tts_engine.create_volume_from_phoneme_and_f0(
         doremi_srore, phonemes, f0s, StyleId(1)
     )
     # Tests
@@ -298,10 +301,10 @@ def test_mocked_synthesize_wave_from_score_output(
     `TTSEngine.frame_synthesize_wave()` の出力スナップショットが一定である
     """
     # Inputs
-    tts_engine = TTSEngine(MockCoreWrapper())
+    tts_engine = SongEngine(MockCoreWrapper())
     doremi_srore = _gen_doremi_score()
     # Outputs
-    result = tts_engine.create_sing_phoneme_and_f0_and_volume(doremi_srore, StyleId(1))
+    result = tts_engine.create_phoneme_and_f0_and_volume(doremi_srore, StyleId(1))
     # Tests
     assert snapshot_json(name="query") == round_floats(
         pydantic_to_native_type(result), round_value=2

test/unit/tts_pipeline/test_tts_engines.py

@@ -45,7 +45,7 @@ def test_tts_engines_get_engine_existing() -> None:
     # Expects
     true_acquired_tts_engine = tts_engine2
     # Outputs
-    acquired_tts_engine = tts_engines.get_engine("0.0.2")
+    acquired_tts_engine = tts_engines.get_tts_engine("0.0.2")
 
     # Test
     assert true_acquired_tts_engine == acquired_tts_engine
@@ -64,7 +64,7 @@ def test_tts_engines_get_engine_latest() -> None:
     # Expects
     true_acquired_tts_engine = tts_engine3
     # Outputs
-    acquired_tts_engine = tts_engines.get_engine(LATEST_VERSION)
+    acquired_tts_engine = tts_engines.get_tts_engine(LATEST_VERSION)
 
     # Test
     assert true_acquired_tts_engine == acquired_tts_engine
@@ -80,4 +80,4 @@ def test_tts_engines_get_engine_missing() -> None:
     tts_engines.register_engine(tts_engine2, "0.0.2")
     # Test
     with pytest.raises(TTSEngineNotFound):
-        tts_engines.get_engine("0.0.3")
+        tts_engines.get_tts_engine("0.0.3")

test/unit/tts_pipeline/test_wave_synthesizer.py

@@ -3,18 +3,20 @@
 import numpy as np
 
 from voicevox_engine.model import AudioQuery
+from voicevox_engine.tts_pipeline.audio_postprocessing import (
+    _apply_output_sampling_rate,
+    _apply_output_stereo,
+    _apply_volume_scale,
+    raw_wave_to_output_wave,
+)
 from voicevox_engine.tts_pipeline.model import AccentPhrase
 from voicevox_engine.tts_pipeline.tts_engine import (
     _apply_intonation_scale,
-    _apply_output_sampling_rate,
-    _apply_output_stereo,
     _apply_pitch_scale,
     _apply_prepost_silence,
     _apply_speed_scale,
-    _apply_volume_scale,
     _count_frame_per_unit,
     _query_to_decoder_feature,
-    raw_wave_to_output_wave,
 )
 
 from .tts_utils import gen_mora, sec

tools/make_docs.py

@@ -10,6 +10,7 @@
 from voicevox_engine.library.library_manager import LibraryManager
 from voicevox_engine.preset.preset_manager import PresetManager
 from voicevox_engine.setting.setting_manager import USER_SETTING_PATH, SettingHandler
+from voicevox_engine.tts_pipeline.song_engine import SongEngineManager
 from voicevox_engine.tts_pipeline.tts_engine import TTSEngineManager
 from voicevox_engine.user_dict.user_dict_manager import UserDictionary
 from voicevox_engine.utility.path_utility import engine_manifest_path, get_save_dir
@@ -39,6 +40,7 @@ def generate_api_docs_html(schema: str) -> str:
     core_manager = CoreManager()
     core_manager.register_core(CoreAdapter(MockCoreWrapper()), "mock")
     tts_engines = TTSEngineManager()
+    song_engines = SongEngineManager()
     tts_engines.register_engine(MockTTSEngine(), "mock")
     preset_path = get_save_dir() / "presets.yaml"
     engine_manifest = load_manifest(engine_manifest_path())
@@ -53,6 +55,7 @@ def generate_api_docs_html(schema: str) -> str:
     # FastAPI の機能を用いて OpenAPI schema を生成する
     app = generate_app(
         tts_engines=tts_engines,
+        song_engines=song_engines,
         core_manager=core_manager,
         setting_loader=SettingHandler(USER_SETTING_PATH),
         preset_manager=PresetManager(preset_path),

voicevox_engine/app/application.py

@@ -31,6 +31,7 @@
 from voicevox_engine.resource_manager import ResourceManager
 from voicevox_engine.setting.model import CorsPolicyMode
 from voicevox_engine.setting.setting_manager import SettingHandler
+from voicevox_engine.tts_pipeline.song_engine import SongEngineManager
 from voicevox_engine.tts_pipeline.tts_engine import TTSEngineManager
 from voicevox_engine.user_dict.user_dict_manager import UserDictionary
 from voicevox_engine.utility.path_utility import engine_root
@@ -39,6 +40,7 @@
 
 def generate_app(
     tts_engines: TTSEngineManager,
+    song_engines: SongEngineManager,
     core_manager: CoreManager,
     setting_loader: SettingHandler,
     preset_manager: PresetManager,
@@ -85,7 +87,9 @@ def _get_core_characters(version: str | None) -> list[CoreCharacter]:
     )
 
     app.include_router(
-        generate_tts_pipeline_router(tts_engines, preset_manager, cancellable_engine)
+        generate_tts_pipeline_router(
+            tts_engines, song_engines, preset_manager, cancellable_engine
+        )
     )
     app.include_router(generate_morphing_router(tts_engines, metas_store))
     app.include_router(

voicevox_engine/app/routers/morphing.py

@@ -87,7 +87,7 @@ def _synthesis_morphing(
         モーフィングの割合は`morph_rate`で指定でき、0.0でベースのスタイル、1.0でターゲットのスタイルに近づきます。
         """
         version = core_version or LATEST_VERSION
-        engine = tts_engines.get_engine(version)
+        engine = tts_engines.get_tts_engine(version)
 
         # モーフィングが許可されないキャラクターペアを拒否する
         characters = metas_store.characters(core_version)