Upload 2 files

handler.py

@@ -0,0 +1,34 @@
+from typing import Dict
+from faster_whisper import WhisperModel
+import io
+
+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
+        # Change to 'cuda' to use the GPU, and set compute_type for faster computation
+        self.model = WhisperModel(model_size, device="cuda", compute_type="float16")
+
+    def __call__(self, data: Dict) -> Dict[str, str]:
+        # Process the input data expected to be in 'inputs' key containing audio file bytes
+        audio_bytes = data["inputs"]
+        
+        # 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)
+        
+        # Compile the results into a text string and extract language information
+        text = " ".join(segment.text for segment in segments)
+        language_code = info.language
+        language_prob = info.language_probability
+
+        # Compile the response dictionary
+        result = {
+            "text": text,
+            "language": language_code,
+            "language_probability": language_prob
+        }
+        
+        return result

requirements.txt

@@ -0,0 +1 @@
+faster-whisper @ https://github.com/guillaumekln/faster-whisper/archive/a4f1cc8f11433e454c3934442b5e1a4ed5e865c3.tar.gz