Update handler.py

handler.py

@@ -1,11 +1,12 @@
 from typing import Dict
-from faster_whisper import WhisperModel
+from faster_whisper import WhisperModel, Streaming
 import io
+import re
 
 class EndpointHandler:
     def __init__(self, model_dir=None):
         # Set model size, assuming installation has been done with appropriate model files and setup
-        model_size = "medium" if model_dir is None else model_dir
+        model_size = "large-v2" if model_dir is None else model_dir
         # Change to 'cuda' to use the GPU, and set compute_type for faster computation
         self.model = WhisperModel(model_size, device="cuda", compute_type="float16")
 
@@ -16,11 +17,14 @@ class EndpointHandler:
         # Convert bytes to a file-like object
         audio_file = io.BytesIO(audio_bytes)
         
-        # Perform transcription using the model
-        segments, info = self.model.transcribe(audio_file)
+        # Enable VAD and perform transcription using the model with a reduced beam size
+        streaming = Streaming(device="cuda", compute_type="float16", vad=True)
+        segments, info = streaming.transcribe(audio_file, beam_size=1)
         
         # Compile the results into a text string and extract language information
-        text = " ".join(segment.text for segment in segments)
+        # Strip leading and trailing whitespace and replace multiple spaces with a single space
+        text = " ".join(segment.text.strip() for segment in segments)
+        text = re.sub(' +', ' ', text)
         language_code = info.language
         language_prob = info.language_probability