Update app.py

app.py

@@ -10,18 +10,25 @@ import uuid
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
-# Load Whisper model from Hugging Face
+
 @st.cache_resource
 def load_model():
-    model_id = "openai/whisper-large-v2"
-
+    # Use a specific Hindi-optimized Whisper model
+    model_id = "openai/whisper-large-v2"  # or consider a multilingual model
+    
+    # For Hindi, you might want to specify additional parameters
     model = AutoModelForSpeechSeq2Seq.from_pretrained(
-        model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
+        model_id, 
+        torch_dtype=torch_dtype, 
+        low_cpu_mem_usage=True, 
+        use_safetensors=True,
     )
     model.to(device)
 
+    # Use the processor from the same model
     processor = AutoProcessor.from_pretrained(model_id)
 
+    # Create pipeline with language specification
     pipe = pipeline(
         "automatic-speech-recognition",
         model=model,
@@ -29,6 +36,7 @@ def load_model():
         feature_extractor=processor.feature_extractor,
         torch_dtype=torch_dtype,
         device=device,
+        generate_kwargs={"language": "hi"}  # Specify Hindi language
     )
     return pipe, processor