Merge branch 'master' into refactor/core_in_metas

Dockerfile

@@ -233,6 +233,7 @@ ADD ./run.py ./presets.yaml ./engine_manifest.json /opt/voicevox_engine/
 ADD ./resources /opt/voicevox_engine/resources
 ADD ./tools/generate_licenses.py /opt/voicevox_engine/tools/
 ADD ./tools/licenses /opt/voicevox_engine/tools/licenses
+ADD ./tools/generate_filemap.py /opt/voicevox_engine/tools/
 
 # Replace version
 ARG VOICEVOX_ENGINE_VERSION=latest
@@ -259,8 +260,7 @@ RUN <<EOF
 EOF
 
 # Generate filemap.json
-ADD ./tools/generate_filemap.py /tmp/
-RUN gosu user /opt/python/bin/python3 /tmp/generate_filemap.py --target_dir resources/character_info
+RUN /opt/python/bin/python3 /opt/voicevox_engine/tools/generate_filemap.py --target_dir /opt/voicevox_engine/resources/character_info
 
 # Keep this layer separated to use layer cache on download failed in local build
 RUN <<EOF

docs/VOICEVOX音声合成エンジンとの連携.md

@@ -2,6 +2,7 @@
 
 - バージョンが上がっても、`/audio_query`で返ってくる値をそのまま`/synthesis`に POST すれば音声合成できるようにする予定です
   - `AudioQuery`のパラメータは増えますが、なるべくデフォルト値で以前と変わらない音声が生成されるようにします
+  - 以前のバージョンの`AudioQuery`を新しいバージョンの`/synthesis`にそのまま POST できるようにします（後方互換）
 - バージョン 0.7 から音声スタイルが実装されました。スタイルの情報は`/speakers`から取得できます
   - スタイルの情報にある`style_id`を`speaker`に指定することで、今まで通り音声合成ができます
     - style_id の指定先が speaker なのは互換性のためです

engine_manifest.json

@@ -44,6 +44,11 @@
             "value": true,
             "name": "全体の音量の調整"
         },
+        "adjust_pause_length": {
+            "type": "bool",
+            "value": true,
+            "name": "句読点などの無音時間の調整"
+        },
         "interrogative_upspeak": {
             "type": "bool",
             "value": true,

run.py

@@ -385,10 +385,10 @@ def main() -> None:
         disable_mutable_api = envs.disable_mutable_api
 
     root_dir = select_first_not_none([args.voicevox_dir, engine_root()])
-    speaker_info_dir = root_dir / "resources" / "character_info"
+    character_info_dir = root_dir / "resources" / "character_info"
     # NOTE: ENGINE v0.19 以前向けに後方互換性を確保する
-    if not speaker_info_dir.exists():
-        speaker_info_dir = root_dir / "speaker_info"
+    if not character_info_dir.exists():
+        character_info_dir = root_dir / "speaker_info"
 
     # ASGI に準拠した VOICEVOX ENGINE アプリケーションを生成する
     app = generate_app(
@@ -400,7 +400,7 @@ def main() -> None:
         engine_manifest,
         library_manager,
         cancellable_engine,
-        speaker_info_dir,
+        character_info_dir,
         cors_policy_mode,
         allow_origin,
         disable_mutable_api=disable_mutable_api,

test/benchmark/engine_preparation.py

@@ -39,7 +39,7 @@ def _generate_engine_fake_server(root_dir: Path) -> TestClient:
         core_manager=core_manager,
         setting_loader=setting_loader,
         preset_manager=preset_manager,
-        speaker_info_dir=root_dir / "resources" / "character_info",
+        character_info_dir=root_dir / "resources" / "character_info",
         user_dict=user_dict,
         engine_manifest=engine_manifest,
         library_manager=library_manager,

test/benchmark/speed/speaker.py

@@ -29,19 +29,19 @@ def benchmark_get_speaker_info_all(
     # speaker_uuid 一覧を準備
     response = client.get("/speakers", params={})
     assert response.status_code == 200
-    speakers = response.json()
-    speaker_uuids = list(map(lambda speaker: speaker["speaker_uuid"], speakers))
+    talk_characters = response.json()
+    uuids = list(map(lambda c: c["speaker_uuid"], talk_characters))
 
     def execute() -> None:
         """計測対象となる処理を実行する"""
-        for speaker_uuid in speaker_uuids:
-            client.get("/speaker_info", params={"speaker_uuid": speaker_uuid})
+        for uuid in uuids:
+            client.get("/speaker_info", params={"speaker_uuid": uuid})
 
     average_time = benchmark_time(execute, n_repeat=10)
     return average_time
 
 
-def benchmark_request_time_for_all_speakers(
+def benchmark_request_time_for_all_talk_characters(
     server: ServerType, root_dir: Path | None = None
 ) -> float:
     """
@@ -54,12 +54,12 @@ def benchmark_request_time_for_all_speakers(
     # speaker_uuid 一覧を準備
     response = client.get("/speakers", params={})
     assert response.status_code == 200
-    speakers = response.json()
-    speaker_uuids = list(map(lambda speaker: speaker["speaker_uuid"], speakers))
+    talk_characters = response.json()
+    uuids = list(map(lambda c: c["speaker_uuid"], talk_characters))
 
     def execute() -> None:
         """計測対象となる処理を実行する"""
-        for _ in speaker_uuids:
+        for _ in uuids:
             client.get("/", params={})
 
     average_time = benchmark_time(execute, n_repeat=10)
@@ -89,7 +89,9 @@ def execute() -> None:
     print(f"全話者 `GET /speaker_info` fakeserve: {result_spk_infos_fakeserve} sec")
     print(f"全話者 `GET /speaker_info` localhost: {result_spk_infos_localhost} sec")
 
-    req_time_all_fake = benchmark_request_time_for_all_speakers("fake", root_dir)
-    req_time_all_local = benchmark_request_time_for_all_speakers("localhost", root_dir)
+    req_time_all_fake = benchmark_request_time_for_all_talk_characters("fake", root_dir)
+    req_time_all_local = benchmark_request_time_for_all_talk_characters(
+        "localhost", root_dir
+    )
     print("全話者 `GET /` fakeserve: {:.3f} sec".format(req_time_all_fake))
     print("全話者 `GET /` localhost: {:.3f} sec".format(req_time_all_local))

test/e2e/test_characters.py

@@ -1,7 +1,4 @@
-"""
-話者・歌手のテスト。
-TODO: 話者と歌手の両ドメイン共通のドメイン用語を定め、このテストファイル名を変更する。
-"""
+"""話者・歌手のテスト"""
 
 import hashlib
 from test.utility import hash_long_string
@@ -42,13 +39,13 @@ def test_話者一覧が取得できる(
 def test_話者の情報を取得できる(
     client: TestClient, snapshot_json: SnapshotAssertion
 ) -> None:
-    speakers = _speaker_list_adapter.validate_python(client.get("/speakers").json())
-    for speaker in speakers:
+    talkers = _speaker_list_adapter.validate_python(client.get("/speakers").json())
+    for talker in talkers:
         response = client.get(
-            "/speaker_info", params={"speaker_uuid": speaker.speaker_uuid}
+            "/speaker_info", params={"speaker_uuid": talker.speaker_uuid}
         )
         assert snapshot_json(
-            name=speaker.speaker_uuid,
+            name=talker.speaker_uuid,
         ) == hash_long_string(response.json())
 
 

test/e2e/test_missing_core.py

@@ -6,6 +6,6 @@
 
 def test_missing_core_422(client: TestClient, snapshot_json: SnapshotAssertion) -> None:
     """存在しないコアを指定するとエラーを返す。"""
-    response = client.get("/supported_devices", params={"core_version": "4.0.4"})
+    response = client.get("/speakers", params={"core_version": "4.0.4"})
     assert response.status_code == 422
     assert snapshot_json == response.json()

test/unit/library/test_library_manager.py

@@ -65,8 +65,8 @@ def setUp(self) -> None:
             self.vvlib_manifest = json.loads(f.read())
             self.library_uuid = self.vvlib_manifest["uuid"]
         with ZipFile(self.library_filename, "w") as zf:
-            speaker_infos = glob.glob("resources/character_info/**", recursive=True)
-            for info in speaker_infos:
+            character_infos = glob.glob("resources/character_info/**", recursive=True)
+            for info in character_infos:
                 zf.write(info)
             zf.writestr(VVLIB_MANIFEST_NAME, json.dumps(self.vvlib_manifest))
         self.library_file = open(self.library_filename, "br")

test/unit/test_metas_store.py

@@ -1,130 +1,96 @@
 import uuid
 
-from voicevox_engine.metas.Metas import Speaker, SpeakerStyle, StyleId, StyleType
+from voicevox_engine.metas.Metas import (
+    SpeakerStyle,
+    SpeakerSupportedFeatures,
+    StyleId,
+    StyleType,
+)
 from voicevox_engine.metas.MetasStore import (
-    SING_STYLE_TYPES,
-    TALK_STYLE_TYPES,
+    _SING_STYLE_TYPES,
+    _TALK_STYLE_TYPES,
     Character,
     filter_characters_and_styles,
 )
 
 
-def _speakers_to_characters(speakers: list[Speaker]) -> list[Character]:
-    """Speaker 配列をキャラクター配列へキャストする。"""
-    characters: list[Character] = []
-    for speaker in speakers:
-        styles = speaker.styles
-        talk_styles = filter(lambda style: style.type in TALK_STYLE_TYPES, styles)
-        sing_styles = filter(lambda style: style.type in SING_STYLE_TYPES, styles)
-        characters.append(
-            Character(
-                name=speaker.name,
-                uuid=speaker.speaker_uuid,
-                talk_styles=list(talk_styles),
-                sing_styles=list(sing_styles),
-                version=speaker.version,
-                supported_features=speaker.supported_features,
-            )
-        )
-    return characters
-
-
-def _gen_speaker(style_types: list[StyleType]) -> Speaker:
-    return Speaker(
-        speaker_uuid=str(uuid.uuid4()),
+def _gen_character(style_types: list[StyleType]) -> Character:
+    talk_styles = list(filter(lambda s: s in _TALK_STYLE_TYPES, style_types))
+    sing_styles = list(filter(lambda s: s in _SING_STYLE_TYPES, style_types))
+    return Character(
         name="",
-        styles=[
-            SpeakerStyle(
-                name="",
-                id=StyleId(0),
-                type=style_type,
-            )
-            for style_type in style_types
+        uuid=str(uuid.uuid4()),
+        talk_styles=[
+            SpeakerStyle(name="", id=StyleId(0), type=style_type)
+            for style_type in talk_styles
+        ],
+        sing_styles=[
+            SpeakerStyle(name="", id=StyleId(0), type=style_type)
+            for style_type in sing_styles
         ],
         version="",
+        supported_features=SpeakerSupportedFeatures(),
     )
 
 
-def _equal_speakers(a: list[Speaker], b: list[Speaker]) -> bool:
+def _equal_characters(a: list[Character], b: list[Character]) -> bool:
     if len(a) != len(b):
         return False
     for i in range(len(a)):
-        if a[i].speaker_uuid != b[i].speaker_uuid:
+        if a[i].uuid != b[i].uuid:
             return False
     return True
 
 
-def test_filter_speakers_and_styles_with_speaker() -> None:
+def test_filter_characters_and_styles_with_talk() -> None:
     # Inputs
-    speaker_talk_only = _gen_speaker(["talk"])
-    speaker_singing_teacher_only = _gen_speaker(["singing_teacher"])
-    speaker_frame_decode_only = _gen_speaker(["frame_decode"])
-    speaker_sing_only = _gen_speaker(["sing"])
-    speaker_allstyle = _gen_speaker(["talk", "singing_teacher", "frame_decode", "sing"])
+    talk_only = _gen_character(["talk"])
+    singing_teacher_only = _gen_character(["singing_teacher"])
+    frame_decode_only = _gen_character(["frame_decode"])
+    sing_only = _gen_character(["sing"])
+    allstyle = _gen_character(["talk", "singing_teacher", "frame_decode", "sing"])
 
     # Outputs
     result = filter_characters_and_styles(
-        _speakers_to_characters(
-            [
-                speaker_talk_only,
-                speaker_singing_teacher_only,
-                speaker_frame_decode_only,
-                speaker_sing_only,
-                speaker_allstyle,
-            ]
-        ),
-        "speaker",
+        [talk_only, singing_teacher_only, frame_decode_only, sing_only, allstyle],
+        "talk",
     )
 
     # Tests
     assert len(result) == 2
 
     # 話者だけになっている
-    assert _equal_speakers(result, [speaker_talk_only, speaker_allstyle])
+    assert _equal_characters(result, [talk_only, allstyle])
 
     # スタイルがフィルタリングされている
-    for speaker in result:
-        for style in speaker.styles:
+    for characters in result:
+        for style in characters.talk_styles + characters.sing_styles:
             assert style.type == "talk"
 
 
-def test_filter_speakers_and_styles_with_singer() -> None:
+def test_filter_characters_and_styles_with_sing() -> None:
     # Inputs
-    speaker_talk_only = _gen_speaker(["talk"])
-    speaker_singing_teacher_only = _gen_speaker(["singing_teacher"])
-    speaker_frame_decode_only = _gen_speaker(["frame_decode"])
-    speaker_sing_only = _gen_speaker(["sing"])
-    speaker_allstyle = _gen_speaker(["talk", "singing_teacher", "frame_decode", "sing"])
+    talk_only = _gen_character(["talk"])
+    singing_teacher_only = _gen_character(["singing_teacher"])
+    frame_decode_only = _gen_character(["frame_decode"])
+    sing_only = _gen_character(["sing"])
+    allstyle = _gen_character(["talk", "singing_teacher", "frame_decode", "sing"])
 
     # Outputs
     result = filter_characters_and_styles(
-        _speakers_to_characters(
-            [
-                speaker_talk_only,
-                speaker_singing_teacher_only,
-                speaker_frame_decode_only,
-                speaker_sing_only,
-                speaker_allstyle,
-            ]
-        ),
-        "singer",
+        [talk_only, singing_teacher_only, frame_decode_only, sing_only, allstyle],
+        "sing",
     )
 
     # Tests
     assert len(result) == 4
 
     # 歌手だけになっている
-    assert _equal_speakers(
-        result,
-        [
-            speaker_singing_teacher_only,
-            speaker_frame_decode_only,
-            speaker_sing_only,
-            speaker_allstyle,
-        ],
+    assert _equal_characters(
+        result, [singing_teacher_only, frame_decode_only, sing_only, allstyle]
     )
 
     # スタイルがフィルタリングされている
-    for speaker in result:
-        for style in speaker.styles:
+    for character in result:
+        for style in character.talk_styles + character.sing_styles:
             assert style.type in ["singing_teacher", "frame_decode", "sing"]