feat(e2k): kanalizerに移行

pyproject.toml

@@ -5,9 +5,9 @@ description = "Add your description here"
 readme = "README.md"
 requires-python = "==3.11.9"
 dependencies = [
-  "e2k==0.3.0",
   "fastapi-slim>=0.115.5",
   "jinja2>=3.1.3", # NOTE: required by fastapi.templating.Jinja2Templates (fastapi-slim's unmanaged dependency)
+  "kanalizer>=0.1.1",
   "numpy>=2.2.3",
   "platformdirs>=4.2.0",
   "pydantic>=2.7.3",

requirements-build.txt

@@ -4,11 +4,11 @@ anyio==4.9.0
 cffi==1.17.1
 click==8.1.8
 colorama==0.4.6 ; sys_platform == 'win32'
-e2k==0.3.0
 fastapi-slim==0.115.12
 h11==0.14.0
 idna==3.10
 jinja2==3.1.6
+kanalizer==0.1.1
 macholib==1.16.3 ; sys_platform == 'darwin'
 markupsafe==3.0.2
 numpy==2.2.4

requirements-dev.txt

@@ -14,7 +14,6 @@ cyclonedx-python-lib==9.1.0
 defusedxml==0.7.1
 distlib==0.3.9
 docopt==0.6.2
-e2k==0.3.0
 fastapi-slim==0.115.12
 filelock==3.18.0
 h11==0.14.0
@@ -24,6 +23,7 @@ identify==2.6.9
 idna==3.10
 iniconfig==2.1.0
 jinja2==3.1.6
+kanalizer==0.1.1
 license-expression==30.4.1
 markdown-it-py==3.0.0
 markupsafe==3.0.2

requirements.txt

@@ -3,11 +3,11 @@ anyio==4.9.0
 cffi==1.17.1
 click==8.1.8
 colorama==0.4.6 ; sys_platform == 'win32'
-e2k==0.3.0
 fastapi-slim==0.115.12
 h11==0.14.0
 idna==3.10
 jinja2==3.1.6
+kanalizer==0.1.1
 markupsafe==3.0.2
 numpy==2.2.4
 platformdirs==4.3.7

run.spec

@@ -8,7 +8,6 @@ datas = [
     ('engine_manifest.json', '.'),
     ('licenses.json', '.'),
 ]
-datas += collect_data_files('e2k')
 datas += collect_data_files('pyopenjtalk')
 
 core_model_dir_path = os.environ.get('CORE_MODEL_DIR_PATH')

test/unit/tts_pipeline/test_katakana_english.py

@@ -44,6 +44,6 @@ def test_convert_english_to_katakana() -> None:
 
     pron = convert_english_to_katakana(string)
 
-    expected_pron = "ボイボ"
+    expected_pron = "ヴォイヴォ"
 
     assert expected_pron == pron

tools/generate_licenses.py

@@ -7,6 +7,7 @@
 """
 
 import argparse
+import importlib.metadata
 import json
 import subprocess
 import sys
@@ -93,7 +94,6 @@ def _update_licenses(pip_licenses: list[_PipLicense]) -> list[_License]:
     """pip から取得したライセンス情報の抜けを補完する。"""
     package_to_license_url: dict[str, str] = {
         # "package_name": "https://license.adress.com/v0.0.0/LICENSE.txt",
-        "e2k": "https://raw.githubusercontent.com/Patchethium/e2k/refs/tags/0.3.0/LICENSE",
     }
 
     updated_licenses = []
@@ -304,11 +304,27 @@ def _add_licenses_manually(licenses: list[_License]) -> None:
     ]
 
 
+def _patch_licenses_manually(licenses: list[_License]) -> None:
+    """手動でライセンス情報を修正する。"""
+    for license in licenses:
+        if license.package_name == "kanalizer":
+            # kanalizerのwheelをビルドするときに使ったライブラリの情報を追加する
+            for p in importlib.metadata.files("kanalizer"):  # type: ignore
+                if p.name == "NOTICE.md":
+                    notice_md = Path(p.locate())
+                    break
+            else:
+                raise Exception("kanalizerのNOTICE.mdが見つかりませんでした。")
+            license.license_text += "\n\n"
+            license.license_text += notice_md.read_text(encoding="utf-8")
+
+
 def _generate_licenses() -> list[_License]:
     pip_licenses = _acquire_licenses_of_pip_managed_libraries()
     licenses = _update_licenses(pip_licenses)
     _validate_license_compliance(licenses)
     _add_licenses_manually(licenses)
+    _patch_licenses_manually(licenses)
 
     return licenses
 

voicevox_engine/tts_pipeline/katakana_english.py

@@ -3,21 +3,11 @@
 import re
 from typing import NewType, TypeGuard
 
-import e2k
+import kanalizer
 
 # 半角アルファベット文字列を示す型
 HankakuAlphabet = NewType("HankakuAlphabet", str)
 
-_global_c2k: e2k.C2K | None = None
-
-
-def _initialize_c2k() -> e2k.C2K:
-    global _global_c2k
-    if _global_c2k is None:
-        _global_c2k = e2k.C2K()
-
-    return _global_c2k
-
 
 # OpenJTalkで使っているアルファベット→カタカナの対応表
 # https://github.com/VOICEVOX/open_jtalk/blob/b9b1bf6a0cba6bc9550b4521913b20334a218dfc/src/njd_set_pronunciation/njd_set_pronunciation_rule_utf_8.h#L397
@@ -69,19 +59,17 @@ def should_convert_english_to_katakana(string: HankakuAlphabet) -> bool:
 
 
 def convert_english_to_katakana(string: HankakuAlphabet) -> str:
-    """e2kを用いて、読みが不明な英単語をカタカナに変換する"""
-    c2k = _initialize_c2k()
-
+    """kanalizerを用いて、読みが不明な英単語をカタカナに変換する"""
     kana = ""
     # キャメルケース的な単語に対応させるため、大文字で分割する
     for word in re.findall("[a-zA-Z][a-z]*", string):
         word = HankakuAlphabet(word)
 
-        # 大文字のみ、もしくは短いワードの場合は、e2kでの解析を行わない
+        # 大文字のみ、もしくは短いワードの場合は、kanalizerでの変換を行わない
         if not should_convert_english_to_katakana(word):
             for alphabet in word:
                 kana += ojt_alphabet_kana_mapping[alphabet.upper()]
         else:
-            kana += c2k(word.lower())
+            kana += kanalizer.convert(word.lower())
 
     return kana