sane-tag-music: use pykakasi for better romanization

pkgs/additional/sane-scripts/default.nix

@@ -205,7 +205,7 @@ let
     tag-music = static-nix-shell.mkPython3 {
       pname = "sane-tag-music";
       srcRoot = ./src;
-      pkgs = [ "python3.pkgs.mutagen" "python3.pkgs.pyexiftool" "python3.pkgs.unidecode" ];
+      pkgs = [ "python3.pkgs.mutagen" "python3.pkgs.pyexiftool" "python3.pkgs.pykakasi" "python3.pkgs.unidecode" ];
     };
     vpn = static-nix-shell.mkBash {
       pname = "sane-vpn";

pkgs/additional/sane-scripts/src/sane-tag-music

@@ -1,5 +1,5 @@
 #!/usr/bin/env nix-shell
-#!nix-shell -i python3 -p python3 -p python3.pkgs.mutagen -p python3.pkgs.pyexiftool -p python3.pkgs.unidecode
+#!nix-shell -i python3 -p python3 -p python3.pkgs.mutagen -p python3.pkgs.pyexiftool -p python3.pkgs.pykakasi -p python3.pkgs.unidecode
 # vim: set filetype=python :
 #
 # standard tags:
@@ -88,27 +88,52 @@ import mutagen.flac
 import mutagen.mp3
 import mutagen.oggopus
 import mutagen.oggvorbis
+import pykakasi
 
 logger = logging.getLogger(__name__)
 
+kks = pykakasi.kakasi()
+
 class MediaType(Enum):
     Audio = "audio"
     Image = "image"
 
+def maybe_romanize(a: str) -> str|None:
+    if a == "かめりあ": return "Camellia"
+    if a == "お握り": return "onigiri"
+    if a == "存流": return "ARU"
+    if a.lower() == "жужжалка": return "zhuzhzhalka"  # Russian
+    if a == "+": return "."
+    if a == "&": return "and"
+
+    # jp kanji -> romanization
+    # pykakasi does word breaking, only return the conversion if on an actual word boundary
+    romaji = [r['hepburn'] for r in kks.convert(a)]
+    if len(romaji) == 1 and romaji[0] != "" and romaji[0] != a:
+        return romaji[0]
+
+    # catchall/fallback
+    if len(a) == 1: return unidecode(a).strip()
+
 def romanize(a: str) -> str:
     """
     transform `a` in a way which loses only a minimal amount of info
     """
-    # ampersand, like: PLS&TY; &I
-    a = a.replace("&", "And")
-    a = a.replace("かめりあ", "Camellia")  # else `unidecode` sets it to kameria
-    a = unidecode(a)
-    # these diacritic replacements might be unnecessary now that i'm using unidecode?
-    a = a.replace("é", "e")
-    a = a.replace("ä", "a")
-    a = a.replace("ö", "o")
-    a = a.replace("ü", "u")
-    return a
+    romanized = ""
+    index = 0
+    while index < len(a):
+        for segment_len in range(10, 0, -1):
+            r = maybe_romanize(a[index : index+segment_len])
+            if r is not None:
+                if len(r) > 1:
+                    r = r[0].upper() + r[1:]
+                if romanized and len(r) != 1:
+                    romanized += " "
+                romanized += r
+                if index + segment_len < len(a) and len(r) != 1:
+                    romanized += " "
+                index += segment_len
+    return romanized
 
 def clean_for_loose_compare(a: str) -> str:
     a = romanize(a)