From faff751faa5b65cd91fa1971901c4dccc360e522 Mon Sep 17 00:00:00 2001
From: pengzhendong <275331498@qq.com>
Date: Tue, 9 Jun 2026 19:35:50 +0800
Subject: [PATCH 1/2] feat: support nbest output in normalize/tag (#296)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

normalize(input, nbest=1) returns a single string (default, backward
compatible). normalize(input, nbest=N) returns a list of N-best results.

Example:
  n.normalize('海淀区108号', nbest=3)
  # ['海淀区一百零八号', '海淀区十八号', '海淀区幺零八号']
---
 tn/processor.py | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)

diff --git a/tn/processor.py b/tn/processor.py
index faabb07..a125b67 100644
--- a/tn/processor.py
+++ b/tn/processor.py
@@ -104,21 +104,29 @@ def build_fst(self, prefix, cache_dir, overwrite_cache):
             logger.info("fst path: {}".format(tagger_path))
             logger.info("          {}".format(verbalizer_path))
 
-    def tag(self, input):
+    def tag(self, input, nbest=1):
         if len(input) == 0:
-            return ""
+            return "" if nbest == 1 else [""]
         input = escape(input)
         lattice = input @ self.tagger
-        return shortestpath(lattice, nshortest=1, unique=True).string()
+        if nbest == 1:
+            return shortestpath(lattice, nshortest=1, unique=True).string()
+        lattice = shortestpath(lattice.project("output").rmepsilon(), nshortest=nbest, unique=True)
+        paths = lattice.paths()
+        results = []
+        while not paths.done():
+            results.append(paths.ostring())
+            paths.next()
+        return results
 
     def verbalize(self, input):
-        # Only words from the blacklist are contained.
         if len(input) == 0:
             return ""
         output = TokenParser(self.ordertype).reorder(input)
-        # We need escape for pynini to build the fst from string.
         lattice = escape(output) @ self.verbalizer
         return shortestpath(lattice, nshortest=1, unique=True).string()
 
-    def normalize(self, input):
-        return self.verbalize(self.tag(input))
+    def normalize(self, input, nbest=1):
+        if nbest == 1:
+            return self.verbalize(self.tag(input))
+        return [self.verbalize(tagged) for tagged in self.tag(input, nbest)]

From e57b889bdd8c896175d2f149d803210df873c8fa Mon Sep 17 00:00:00 2001
From: pengzhendong <275331498@qq.com>
Date: Tue, 9 Jun 2026 19:53:15 +0800
Subject: [PATCH 2/2] fix: only use digit-by-digit reading for 4-digit years
 (#302)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Previously 2-digit numbers followed by 年/年度/赛季 were read
digit-by-digit (38年 => 三八年). Now only 4-digit years use this
pattern, so 38年 => 三十八年 (via number) while 2024年 => 二零二四年.
---
 tn/chinese/rules/measure.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tn/chinese/rules/measure.py b/tn/chinese/rules/measure.py
index f5ecb4a..cac5bf1 100644
--- a/tn/chinese/rules/measure.py
+++ b/tn/chinese/rules/measure.py
@@ -47,12 +47,12 @@ def build_tagger(self):
             measure @= self.build_rule(cross("两" + unit, "二" + unit), l="[BOS]")
             measure @= self.build_rule(cross("到两" + unit, "到二" + unit), r="[EOS]")
 
-        # -xxxx年, -xx年
+        # xxxx年, xxxx-xxxx年
         digits = self.cardinal.digits
-        cardinal = digits**2 | digits**4
+        yyyy = digits**4
         unit = accep("年") | accep("年度") | accep("赛季")
-        prefix = cardinal + (rmspace + unit).ques + to
-        annual = prefix.ques + cardinal + unit
+        prefix = yyyy + (rmspace + unit).ques + to
+        annual = prefix.ques + yyyy + unit
 
         tagger = insert('value: "') + (measure | annual) + insert('"')