refactor: replace Kokoro-82M with suno/bark for TTS, update HF cache path, and add model warmup on startup.

backend/main.py

@@ -1,4 +1,4 @@
-import asyncio
+import asyncio
 import base64
 import logging
 import os
@@ -37,6 +37,15 @@ def _get_markitdown():
 
 app = FastAPI()
 
+@app.on_event("startup")
+async def startup_event():
+    logger.info("Starting blocking preload for TTS and ASR models...")
+    try:
+        from tts_asr import _warmup_all
+        await _warmup_all()
+    except Exception as e:
+        logger.warning(f"Failed to initiate model warmup: {e}")
+
 ACTIVE_COMPLETIONS: dict[str, asyncio.Task] = {}
 ACTIVE_COMPLETIONS_LOCK = asyncio.Lock()