Add substitutions option in ASR evaluation

examples/asr/speech_to_text_eval.py

@@ -105,6 +105,7 @@ class EvaluationConfig(transcribe_speech.TranscriptionConfig):
             separate_punctuation=False,
             do_lowercase=False,
             rm_punctuation=False,
+            substitutions="",
         )
     )
 
@@ -154,7 +155,7 @@ def main(cfg: EvaluationConfig):
 
             predicted_text.append(data["pred_text"])
 
-    pc = PunctuationCapitalization(cfg.text_processing.punctuation_marks)
+    pc = PunctuationCapitalization(cfg.text_processing.punctuation_marks, cfg.text_processing.substitutions)
     if cfg.text_processing.separate_punctuation:
         ground_truth_text = pc.separate_punctuation(ground_truth_text)
         predicted_text = pc.separate_punctuation(predicted_text)
@@ -164,6 +165,9 @@ def main(cfg: EvaluationConfig):
     if cfg.text_processing.rm_punctuation:
         ground_truth_text = pc.rm_punctuation(ground_truth_text)
         predicted_text = pc.rm_punctuation(predicted_text)
+    if cfg.text_processing.substitutions:
+        ground_truth_text = pc.substitute_equivalents(ground_truth_text)
+        predicted_text = pc.substitute_equivalents(predicted_text)
 
     # Test for invalid manifest supplied
     if invalid_manifest:

nemo/collections/asr/parts/utils/transcribe_utils.py

@@ -637,20 +637,27 @@ def compute_metrics_per_sample(
 
 
 class PunctuationCapitalization:
-    def __init__(self, punctuation_marks: str):
+    def __init__(self, punctuation_marks: str, substitutions: str):
         """
         Class for text processing with punctuation and capitalization. Can be used with class TextProcessingConfig.
 
         Args:
             punctuation_marks (str): String with punctuation marks to process.
+            substitutions (str): String of equivalencies to substitute, separated with `;`.
         Example: punctuation_marks = '.,?'
+        substitutions = "fi~fi"
         """
         if punctuation_marks:
             self.regex_punctuation = re.compile(fr"([{''.join(punctuation_marks)}])")
             self.regex_extra_space = re.compile(r'\s{2,}')
         else:
             self.regex_punctuation = None
 
+        if substitutions:
+            self.substitutions = self._parse_substitutions(substitutions)
+        else:
+            self.substitutions = None
+
     def separate_punctuation(self, lines: List[str]) -> List[str]:
         if self.regex_punctuation is not None:
             return [
@@ -668,6 +675,35 @@ def rm_punctuation(self, lines: List[str]) -> List[str]:
         else:
             return lines
 
+    def substitute_equivalents(self, lines: List[str]) -> List[str]:
+        if self.substitutions is not None:
+            return [line.replace(orig, sub) for orig, sub in self.substitutions.items() for line in lines]
+        else:
+            return lines
+
+    @staticmethod
+    def _parse_substitutions(s: str) -> dict[str, str]:
+        """
+        Parse substitutions from a string: "src~dst;src2~dst2;..."
+
+        Supports either literal Unicode (preferred) or escape sequences like "\\u0587".
+        """
+        # Decode \uXXXX / \UXXXXXXXX only when present (avoid surprising behavior).
+        decode = lambda t: t.encode("utf-8").decode("unicode_escape") if ("\\u" in t or "\\U" in t) else t
+
+        subs = {}
+        for raw in (p.strip() for p in s.split(";")):
+            if not raw:
+                continue
+            if "~" not in raw:
+                raise ValueError(f"Invalid substitution '{raw}'. Expected 'SRC~DST'.")
+            src, dst = (x.strip() for x in raw.split("~", 1))
+            if not src:
+                raise ValueError(f"Invalid substitution '{raw}'. SRC must be non-empty.")
+            subs[decode(src)] = decode(dst)
+
+        return subs
+
 
 @dataclass
 class TextProcessingConfig:
@@ -682,3 +718,6 @@ class TextProcessingConfig:
 
     # Whether to separate punctuation with the previouse word by space.
     separate_punctuation: bool = True
+
+    # Characters (or combinations of) to treat as equivalent
+    substitutions: str = ""