diff --git a/doc/protocol.rst b/doc/protocol.rst
index 8fde63888..aa2cbb272 100644
--- a/doc/protocol.rst
+++ b/doc/protocol.rst
@@ -237,6 +237,7 @@ of:
 The :command:`find` commands are case sensitive, while
 :command:`search` and related commands ignore case.  The latter also
 applies `Unicode normalization <https://unicode.org/reports/tr15/>`__
+and converts all punctuation to ASCII equivalents
 if MPD was compiled with `ICU <https://icu.unicode.org/>`__ support.
 
 Prior to MPD 0.21, the syntax looked like this::
diff --git a/src/lib/icu/Canonicalize.cxx b/src/lib/icu/Canonicalize.cxx
index 08528c592..aca407584 100644
--- a/src/lib/icu/Canonicalize.cxx
+++ b/src/lib/icu/Canonicalize.cxx
@@ -26,15 +26,48 @@
 
 #ifdef HAVE_ICU
 #include "Normalize.hxx"
+#include "Transliterator.hxx"
 #include "Util.hxx"
 #include "util/AllocatedArray.hxx"
 #include "util/SpanCast.hxx"
 #endif
 
+#ifdef HAVE_ICU
+
+using std::string_view_literals::operator""sv;
+
+static IcuTransliterator *transliterator;
+
+void
+IcuCanonicalizeInit()
+{
+	assert(transliterator == nullptr);
+
+	const auto id =
+		/* convert all punctuation to ASCII equivalents */
+		"[:Punctuation:] Latin-ASCII;"sv;
+
+	transliterator = new IcuTransliterator(ToStringView(std::span{UCharFromUTF8(id)}),
+					       {});
+}
+
+void
+IcuCanonicalizeFinish() noexcept
+{
+	assert(transliterator != nullptr);
+
+	delete transliterator;
+	transliterator = nullptr;
+}
+
+#endif
+
 AllocatedString
 IcuCanonicalize(std::string_view src, bool fold_case) noexcept
 try {
 #ifdef HAVE_ICU
+	assert(transliterator != nullptr);
+
 	auto u = UCharFromUTF8(src);
 	if (u.data() == nullptr)
 		return {src};
@@ -45,6 +78,10 @@ try {
 	    n != nullptr)
 		u = std::move(n);
 
+	if (auto t = transliterator->Transliterate(ToStringView(std::span{u}));
+	    t != nullptr)
+		u = std::move(t);
+
 	return UCharToUTF8(ToStringView(std::span{u}));
 #else
 #error not implemented
diff --git a/src/lib/icu/Canonicalize.hxx b/src/lib/icu/Canonicalize.hxx
index d2d33c004..33c0f8202 100644
--- a/src/lib/icu/Canonicalize.hxx
+++ b/src/lib/icu/Canonicalize.hxx
@@ -28,6 +28,15 @@
 
 class AllocatedString;
 
+/**
+ * Throws on error.
+ */
+void
+IcuCanonicalizeInit();
+
+void
+IcuCanonicalizeFinish() noexcept;
+
 /**
  * Transform the given string to "canonical" form to allow fuzzy
  * string comparisons.  The full set of features (if ICU is being
diff --git a/src/lib/icu/Init.cxx b/src/lib/icu/Init.cxx
index 305b7a39a..dda779597 100644
--- a/src/lib/icu/Init.cxx
+++ b/src/lib/icu/Init.cxx
@@ -19,6 +19,7 @@
 
 #include "Init.hxx"
 #include "Collate.hxx"
+#include "Canonicalize.hxx"
 #include "util/RuntimeError.hxx"
 
 #include <unicode/uclean.h>
@@ -33,11 +34,13 @@ IcuInit()
 					 u_errorName(code));
 
 	IcuCollateInit();
+	IcuCanonicalizeInit();
 }
 
 void
 IcuFinish() noexcept
 {
+	IcuCanonicalizeFinish();
 	IcuCollateFinish();
 
 	u_cleanup();
diff --git a/test/TestStringFilter.cxx b/test/TestStringFilter.cxx
index 4b526759b..a09bf1884 100644
--- a/test/TestStringFilter.cxx
+++ b/test/TestStringFilter.cxx
@@ -143,6 +143,18 @@ TEST_F(StringFilterTest, Normalize)
 
 #endif
 
+#ifdef HAVE_ICU
+
+TEST_F(StringFilterTest, Transliterate)
+{
+	const StringFilter f{"'", true, false, false, false};
+
+	EXPECT_TRUE(f.Match("’"));
+	EXPECT_FALSE(f.Match("\""));
+}
+
+#endif
+
 TEST_F(StringFilterTest, FoldCase)
 {
 	const StringFilter f{"nëedlé", true, false, false, false};