Update app.py

app.py

@@ -6,6 +6,7 @@ import requests
 import torch
 import whisperx
 import gc
+import pandas as pd
 from datetime import timedelta
 
 # --- Configuration & Tokens ---
@@ -90,8 +91,8 @@ with st.sidebar:
     fps = st.number_input("Timeline FPS", value=25)
     
     st.header("Model Settings")
-    model_size = st.selectbox("Whisper Model Size", ["large-v3", "large-v2", "medium", "base"], index=0)
-    num_speakers = st.number_input("Number of Speakers (Optional)", min_value=0, value=0, help="Set to 0 for auto-detect")
+    model_size = st.selectbox("Whisper Model", ["large-v2", "medium", "base"], index=0)
+    num_speakers = st.number_input("Speakers (0=Auto)", min_value=0, value=0)
     
     st.divider()
     st.info("API Keys are managed via Environment Secrets.")
@@ -109,10 +110,7 @@ if uploaded_file:
             if not ACTIVE_HF_TOKEN or "PASTE_YOUR_HF_TOKEN" in ACTIVE_HF_TOKEN:
                 st.error("Please provide a valid Hugging Face Token.")
             else:
-                status_container = st.empty()
-                with status_container.container():
-                    st.write("🔄 **Processing Started...**")
-                    
+                with st.spinner("Processing... This may take a moment."):
                     # Save local temp file
                     with open("temp_input", "wb") as f:
                         f.write(uploaded_file.getbuffer())
@@ -129,8 +127,7 @@ if uploaded_file:
                         device = "cuda" if torch.cuda.is_available() else "cpu"
                         st.write(f"🚀 **Loading WhisperX on {device}...**")
                         
-                        # 1. Transcribe with WhisperX (Faster-Whisper)
-                        batch_size = 16 # Reduce if low VRAM
+                        # 1. Transcribe
                         # Use float16 for GPU, int8 for CPU
                         compute_type = "float16" if device == "cuda" else "int8"
                         
@@ -138,28 +135,27 @@ if uploaded_file:
                         
                         st.write("📝 **Transcribing...**")
                         audio = whisperx.load_audio("temp_audio.wav")
-                        result = model.transcribe(audio, batch_size=batch_size)
+                        result = model.transcribe(audio, batch_size=16)
                         
                         # Cleanup VRAM
-                        model_a = None
                         gc.collect()
                         torch.cuda.empty_cache()
+                        del model
                         
-                        # 2. Align (Improves timestamp accuracy)
+                        # 2. Align
                         st.write("⏱️ **Aligning Audio...**")
                         model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)
                         result = whisperx.align(result["segments"], model_a, metadata, audio, device, return_char_alignments=False)
                         
                         # Cleanup VRAM
-                        model_a = None
                         gc.collect()
                         torch.cuda.empty_cache()
+                        del model_a
                         
                         # 3. Diarize
                         st.write("🗣️ **Diarizing Speakers...**")
                         diarize_model = whisperx.DiarizationPipeline(use_auth_token=ACTIVE_HF_TOKEN, device=device)
                         
-                        # Optional: Enforce speaker count
                         diarize_kwargs = {}
                         if num_speakers > 0:
                             diarize_kwargs = {"min_speakers": num_speakers, "max_speakers": num_speakers}