Update app.py

app.py

@@ -8,16 +8,18 @@ import torch
 import torchaudio
 from datetime import timedelta
 from pyannote.audio import Pipeline
-from huggingface_hub import login
+from huggingface_hub import login, hf_hub_download
 from pydub import AudioSegment
 
 # --- Configuration & Tokens ---
+# Hardcode tokens here if you want to avoid UI input
 HARDCODED_HF_TOKEN = "PASTE_YOUR_HF_TOKEN_HERE"
 HARDCODED_GEMINI_KEY = ""
 
 ENV_HF_TOKEN = os.environ.get("HF_TOKEN", "")
 ENV_GEMINI_KEY = os.environ.get("GEMINI_API_KEY", "")
 
+# Determine active keys
 ACTIVE_HF_TOKEN = ENV_HF_TOKEN if ENV_HF_TOKEN else HARDCODED_HF_TOKEN
 ACTIVE_GEMINI_KEY = ENV_GEMINI_KEY if ENV_GEMINI_KEY else HARDCODED_GEMINI_KEY
 
@@ -102,18 +104,20 @@ with st.sidebar:
 uploaded_file = st.file_uploader("Upload Video/Audio Clip", type=["mp4", "m4a", "wav", "mp3", "mov"])
 
 if uploaded_file:
+    # --- Step 1: Technical Processing ---
     if "transcript" not in st.session_state:
         if st.button("Step 1: Transcribe & Diarize"):
             if not ACTIVE_HF_TOKEN or "PASTE_YOUR_HF_TOKEN" in ACTIVE_HF_TOKEN:
                 st.error("Please provide a valid Hugging Face Token.")
             else:
-                with st.spinner("Processing..."):
+                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())
                     
                     st.write("🎵 **Preprocessing Audio...**")
                     try:
-                        # Use PyDub for robust WAV conversion
+                        # Use PyDub to convert to WAV (Mono, 16kHz)
                         audio = AudioSegment.from_file("temp_input")
                         audio = audio.set_channels(1)
                         audio = audio.set_frame_rate(16000)
@@ -126,16 +130,22 @@ if uploaded_file:
                     st.write("🗣️ **Running Speaker Diarization...**")
                     diarization = None
                     try:
-                        # Login with token first
+                        # Log in globally
                         login(token=ACTIVE_HF_TOKEN)
                         
-                        # Load pipeline using legacy API (use_auth_token is valid here)
-                        # NOTE: Using the older model ID for 2.1 compatibility
-                        pipeline = Pipeline.from_pretrained(
-                            "pyannote/speaker-diarization@2.1",
-                            use_auth_token=ACTIVE_HF_TOKEN
+                        # Use manual config download with explicit 'token' argument
+                        # This fixes the hf_hub_download error
+                        config_path = hf_hub_download(
+                            repo_id="pyannote/speaker-diarization", 
+                            revision="2.1",
+                            filename="config.yaml", 
+                            token=ACTIVE_HF_TOKEN
                         )
                         
+                        # Load pipeline from the manually downloaded config
+                        # We do NOT pass a token here because the config file is local
+                        pipeline = Pipeline.from_pretrained(config_path)
+                        
                         if torch.cuda.is_available():
                             st.write("🚀 Using GPU for Diarization")
                             pipeline.to(torch.device("cuda"))
@@ -159,8 +169,8 @@ if uploaded_file:
                     speaker_turns = []
                     
                     if diarization:
-                        # 2.1.1 returns a proper Annotation object directly
                         try:
+                            # 2.1.1 returns a proper Annotation object directly
                             for turn, _, speaker_id in diarization.itertracks(yield_label=True):
                                 speaker_turns.append({"start": turn.start, "end": turn.end, "speaker": speaker_id})
                             
@@ -169,7 +179,7 @@ if uploaded_file:
                             else:
                                 st.warning("⚠️ Pipeline ran but returned no tracks.")
                         except AttributeError:
-                             st.error("Could not iterate tracks.")
+                             st.error("Could not iterate tracks. Output object format mismatch.")
 
                     for segment in result['segments']:
                         mid_time = (segment['start'] + segment['end']) / 2