Add model downloading functionality for Faster Whisper in app.py, enabling efficient local caching and improved model loading performance.

app.py

@@ -48,6 +48,20 @@ try:
 except OSError as e:
     sys.exit(f"❌  Could not load {cnn_so} : {e}")
 
+
+from huggingface_hub import snapshot_download
+MODEL_REPO = "deepdml/faster-whisper-large-v3-turbo-ct2"   # CT2 format
+LOCAL_DIR  = f"{CACHE_ROOT}/whisper_turbo"
+
+# Download once; later runs are instant
+snapshot_download(
+    repo_id=MODEL_REPO,
+    local_dir=LOCAL_DIR,
+    local_dir_use_symlinks=True,   # saves disk space
+    resume_download=True
+)
+model_cache_path = LOCAL_DIR      # <‑‑ this is what we pass to WhisperModel
+
 # Lazy global holder ----------------------------------------------------------
 _whisper = None
 _diarizer = None
@@ -77,7 +91,7 @@ def _load_models():
     if _whisper is None:
         print("Loading Whisper model...")
         _whisper = WhisperModel(
-            "large-v3-turbo",
+            model_cache_path,
             device="cuda",
             compute_type="float16",
         )