fix: websocket issue

websocket.create_connection is short lived, so whenever the inference
server is needed, it needs to be called again to reconnect to the
server.

server.py

@@ -13,17 +13,16 @@ current_index = -1
 t = None
 # websocket connection to the inference server
 ws = None
+ws_url = ""
+
 
 @asynccontextmanager
 async def lifespan(app: FastAPI):
-    global ws
+    global ws, ws_url
 
-    url = os.environ.get("INFERENCE_SERVER_WS_URL")
-    if not url:
-        url = "ws://localhost:8001"
-
-    ws = websocket.create_connection(url)
-    print(f"websocket connected to {url}")
+    ws_url = os.environ.get("INFERENCE_SERVER_WS_URL")
+    if not ws_url:
+        ws_url = "ws://localhost:8001"
 
     advance()
 
@@ -36,7 +35,7 @@ async def lifespan(app: FastAPI):
 
 
 def generate_new_audio():
-    if not ws:
+    if not ws_url:
         return
 
     global current_index
@@ -51,6 +50,9 @@ def generate_new_audio():
 
     print("generating new audio...")
 
+    ws = websocket.create_connection(ws_url)
+    print(f"websocket connected to {ws_url}")
+
     ws.send("generate")
 
     wavs = []
@@ -66,6 +68,8 @@ def generate_new_audio():
 
     print("audio generated.")
 
+    ws.close()
+
 
 def advance():
     global current_index, t