Update app.py

app.py

@@ -1,14 +1,9 @@
 import gradio as gr
-import torch
 import whisper
 import librosa
-from transformers import pipeline
+import numpy as np
 
-# Check if DistilWhisper is available on Hugging Face
-# This is a placeholder model name, update it with an actual distillation model if available
-# distil_whisper_model = "huggingface/distil-whisper-model"
-
-# If no distil version, load smaller Whisper model for speed (e.g., "base" or "tiny")
+# Load Whisper model (using tiny for faster performance)
 model = whisper.load_model("tiny")
 
 # Chunking function to split the audio into smaller parts (e.g., 5-second chunks)
@@ -38,16 +33,15 @@ def transcribe_audio_in_chunks(audio_file):
     if audio_file is None:
         return "No audio file provided."
     
-    # Check the audio file path
-    if not os.path.exists(audio_file):
-        return "The audio file does not exist or is inaccessible."
-    
     # Chunk the audio into 5-second parts
     chunks, sr = chunk_audio(audio_file, chunk_size=5)
 
     # Process each chunk and append the results as real-time transcription
     transcription = ""
     for i, chunk in enumerate(chunks):
+        # Convert the chunk into the correct format for Whisper (numpy array of floats)
+        chunk = np.array(chunk)
+        
         # Transcribe each chunk
         result = model.transcribe(chunk)
         transcription += f"Chunk {i + 1}: {result['text']}\n"