Add punctuation + capitalization formatting layer

app.py

@@ -44,36 +44,8 @@ BASE = "openai/whisper-large-v3-turbo"
 ADAPTER = "michaelodafe/whisper-pidgin-v1"
 MAX_SECONDS = 30
 
-INITIAL_PROMPT = (
-    "dis na bbc news pidgin tori about buhari tinubu atiku saraki "
-    "tony nwoye femi otedola akinwunmi ambode oseloka henry obaze "
-    "zainab balogun jimoh moshood and aisha for nigeria politics "
-    "for states like lagos anambra delta kogi niger abuja kano rivers "
-    "edo ogun salford and offa with organizations like apc pdp nema "
-    "jamb frsc brt jp morgan and wikipedia pipo dey tok say di dey na "
-    "wey pikin tori sabi hapun sometin anytin becos redi alredi neva "
-    "dem una abi oga chillax snakebite"
-)
-
-_DIGIT_PAIR = re.compile(r"(\d) (\d)")
-_PUNCT = re.compile(r"[.,!?;:\"]")
-_INTRA_NUM_COMMA = re.compile(r"(\d),(\d)")
-
-
-def postprocess(text: str) -> str:
-    while True:
-        new = _INTRA_NUM_COMMA.sub(r"\1\2", text)
-        if new == text:
-            break
-        text = new
-    text = _PUNCT.sub("", text)
-    while True:
-        new = _DIGIT_PAIR.sub(r"\1\2", text)
-        if new == text:
-            break
-        text = new
-    return re.sub(r" +", " ", text).strip()
-
+# Shared decode helpers (hotword prompt + punctuation/casing formatter).
+from decode import INITIAL_PROMPT, format_output
 
 print("Loading processor + base model + adapter...")
 processor = WhisperProcessor.from_pretrained(BASE, language="english", task="transcribe")
@@ -111,7 +83,7 @@ def transcribe(audio):
             gen_kwargs["prompt_ids"] = prompt_ids
         output_ids = model.generate(inputs.input_features, **gen_kwargs)
         text = processor.batch_decode(output_ids, skip_special_tokens=True)[0]
-        return postprocess(text.strip().lower())
+        return format_output(text)
     except Exception as e:
         return f"⚠️ Error: {type(e).__name__}: {e}"