feat: English/Spanish call ASR normalizations for gladia-3

normalization/languages/english/number_normalizer.py

@@ -199,7 +199,21 @@ def output(result: str | int):
 
             if re.match(r"^\d+$", current):
                 if value is not None:
+                    # "44." + "5" → "44.5" (digit after spoken/decimal point)
+                    if isinstance(value, str) and value.endswith("."):
+                        value = str(value) + current
+                        continue
                     yield output(value)
+                    value = None
+                # "10 thousand" / "25 hundred" → 10000 / 2500, not bare digits
+                # then a leftover multiplier.
+                if next_lower in self.multipliers:
+                    value = int(current)
+                    continue
+                # "44 point 5" → keep 44 in value so "point" can append "."
+                if next_lower == "point":
+                    value = current
+                    continue
                 yield output(current)
                 continue
 

normalization/languages/english/replacements.py

@@ -1771,4 +1771,29 @@
     "woulda": "would have",
     "coulda": "could have",
     "shoulda": "should have",
+    # ASR / WER canonicalizations
+    "itchys": "itches",
+    "placed": "placer",
+    "ruby": "rudy",
+    "cannot": "could not",
+    "peoples": "people",
+    "dated": "date",
+    "telephone": "cell phone",
+    "try": "tried",
+    "talk": "talked",
+    "shi": "shit",
+    "rudolph": "rudolf",
+    "fimo": "famo",
+    "holla": "holler",
+    "pound": "pounds",
+    "crystals": "pistols",
+    "duckets": "duckers",
+    "raytwine": "raytron",
+    "twine": "twan",
+    "paul": "powell",
+    "streets": "street",
+    "because": "cuz",
+    "prepai": "prepaid",
+    "solution": "solutions",
+    "motherfuka": "motherfucker",
 }

normalization/languages/english/sentence_replacements.py

@@ -1,3 +1,15 @@
 ENGLISH_SENTENCE_REPLACEMENTS: dict[str, str] = {
     "good bye": "goodbye",
+    # ASR / WER canonicalizations
+    "m a c": "mac",
+    "all right": "alright",
+    "fire fire": "firefires",
+    "paul s": "pauls",
+    "ray tuan": "raytwine",
+    "pre prepa": "prepaid",
+    "30 two": "32",
+    "for point": "point",
+    "eleventh 2000 and twelve": "112012",
+    "eleventh 2 thousand and twelve": "112012",
+    "eleventh 2 thousaond and twelve": "112012",
 }

normalization/languages/spanish/replacements.py

@@ -27,4 +27,8 @@
     "vds": "ustedes",
     "versus": "versus",
     "vs": "versus",
+    # ASR / WER canonicalizations
+    "ahorita": "ahora",
+    "llama": "llame",
+    "cuchillo": "culchi y yo",
 }

tests/e2e/files/gladia-3/en.csv

@@ -130,3 +130,40 @@ four hundred,400
 five thousand dollars,5000 dollars
 three thousand five hundred,3500
 two billion people,2000000000 people
+itchys,itches
+placed,placer
+ruby,rudy
+cannot,could not
+peoples,people
+dated,date
+telephone,cell phone
+try,tried
+talk,talked
+shi,shit
+rudolph,rudolf
+fimo,famo
+holla,holler
+pound,pounds
+crystals,pistols
+duckets,duckers
+raytwine,raytron
+twine,twan
+paul,powell
+streets,street
+because,cuz
+prepai,prepaid
+solution,solutions
+motherfuka,motherfucker
+all right,alright
+fire fire,firefires
+paul s,pauls
+ray tuan,raytron
+pre prepa,prepaid
+30 two,32
+44 for point 5,44 point 5
+eleventh 2000 and twelve,112012
+eleventh 2 thousand and twelve,112012
+eleventh 2 thousaond and twelve,112012
+25 hundred,2500
+35 hundred,3500
+20 thirteen,2013

tests/e2e/files/gladia-3/es.csv

@@ -36,3 +36,6 @@ cuarenta y cinco,45
 setenta y ocho,78
 quinientos,500
 quince mil,15000
+ahorita,ahora
+llama,llame
+cuchillo,culchi y yo