add prompt

app.py

@@ -15,6 +15,14 @@ os.environ.update(
     MPLCONFIGDIR      = f"{CACHE_ROOT}/mpl",
 )
 
+INITIAL_PROMPT = '''
+Transcribe the audio verbatim in the original language(s). Do NOT translate or summarize.
+Use the standard punctuation of each language to avoid run-on sentences.
+• End every sentence with a sentence-final mark (., ?, ! or the local equivalent such as 。！？).
+• Insert commas/pauses where they naturally occur; prefer more punctuation over less.
+• For CJK languages, use native full-width marks (，、。！？：；「」『』（ ）); for others, use their customary marks.
+'''
+
 # make sure the directories exist
 for path in os.environ.values():
     pathlib.Path(path).mkdir(parents=True, exist_ok=True)
@@ -384,9 +392,11 @@ class WhisperTranscriber:
             language_detection_segments=1,
             task="translate" if translate else "transcribe",
         )
-        
+        transcribe_model = whisper
+        if batch_size > 1:
+            transcribe_model = batched_whisper
         # Use batched inference for better performance
-        segments, transcript_info = whisper.transcribe(
+        segments, transcript_info = transcribe_model.transcribe(
             audio_path, 
             #batch_size=batch_size, 
             **options