Use writable model cache for diarization models; vendor improved_diarization in src\

src/diarization.py

@@ -19,6 +19,7 @@ from pathlib import Path
 from typing import List, Tuple, Optional, Callable, Dict, Any
 import streamlit as st
 import logging
+from utils import get_writable_model_dir
 
 # Import the improved diarization pipeline (robust: search repo tree)
 try:
@@ -82,12 +83,15 @@ def download_diarization_models():
     Only downloads embedding model - we'll use Silero VAD for segmentation
     Returns tuple (embedding_model_path, success)
     """
-    models_dir = Path("models/diarization")
+    # Use a writable cache directory (works on HF Spaces and local)
+    cache_dir = get_writable_model_dir()
+    models_dir = cache_dir / "diarization"
     models_dir.mkdir(parents=True, exist_ok=True)
-    
+
     # Updated to optimal Chinese/Taiwanese model from benchmark results
     # 3dspeaker_campplus_zh_en_advanced: F1=0.500, 60.5ms, 27MB
     embedding_model = models_dir / "3dspeaker_speech_campplus_sv_zh_en_16k-common_advanced.onnx"
+    st.info(f"Model cache directory: {models_dir}")
     
     try:
         # Check if embedding model exists