refactor: remove conditional voice mode support from backend and frontend (#8597)

* revert changes that made voice mode conditional to the presence of webrtcvad

* refactor: lazy-load webrtcvad dependency

* refactor: remove try-except block and directly import voice_mode_router

* [autofix.ci] apply automated fixes

* feat: add LRU caching to voice activity detector initialization

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

src/backend/base/langflow/api/router.py

@@ -19,6 +19,7 @@ from langflow.api.v1 import (
     validate_router,
     variables_router,
 )
+from langflow.api.v1.voice_mode import router as voice_mode_router
 from langflow.api.v2 import files_router as files_router_v2
 from langflow.api.v2 import mcp_router as mcp_router_v2
 
@@ -45,18 +46,12 @@ router_v1.include_router(folders_router)
 router_v1.include_router(projects_router)
 router_v1.include_router(starter_projects_router)
 router_v1.include_router(mcp_router)
+router_v1.include_router(voice_mode_router)
 router_v1.include_router(mcp_projects_router)
 
 router_v2.include_router(files_router_v2)
 router_v2.include_router(mcp_router_v2)
 
-try:
-    from langflow.api.v1.voice_mode import router as voice_mode_router
-
-    router_v1.include_router(voice_mode_router)
-except ImportError:
-    pass
-
 router = APIRouter(
     prefix="/api",
 )

src/backend/base/langflow/api/utils.py

@@ -378,11 +378,3 @@ async def verify_public_flow_and_get_user(flow_id: uuid.UUID, client_id: str | N
         raise HTTPException(status_code=403, detail=msg)
 
     return user, new_flow_id
-
-
-def get_voice_mode_enabled() -> bool:
-    try:
-        import webrtcvad  # noqa: F401
-    except ImportError:
-        return False
-    return True

src/backend/base/langflow/api/v1/__init__.py

@@ -14,6 +14,7 @@ from langflow.api.v1.store import router as store_router
 from langflow.api.v1.users import router as users_router
 from langflow.api.v1.validate import router as validate_router
 from langflow.api.v1.variable import router as variables_router
+from langflow.api.v1.voice_mode import router as voice_mode_router
 
 __all__ = [
     "api_key_router",
@@ -32,11 +33,5 @@ __all__ = [
     "users_router",
     "validate_router",
     "variables_router",
+    "voice_mode_router",
 ]
-
-try:
-    from langflow.api.v1.voice_mode import router as voice_mode_router
-
-    __all__ += ["voice_mode_router"]
-except ImportError:
-    pass

src/backend/base/langflow/api/v1/endpoints.py

@@ -14,7 +14,7 @@ from fastapi.responses import StreamingResponse
 from loguru import logger
 from sqlmodel import select
 
-from langflow.api.utils import CurrentActiveUser, DbSession, get_voice_mode_enabled, parse_value
+from langflow.api.utils import CurrentActiveUser, DbSession, parse_value
 from langflow.api.v1.schemas import (
     ConfigResponse,
     CustomComponentRequest,
@@ -752,7 +752,6 @@ async def get_config():
         return {
             "feature_flags": FEATURE_FLAGS,
             **settings_service.settings.model_dump(),
-            "voice_mode_enabled": get_voice_mode_enabled(),
         }
     except Exception as exc:
         raise HTTPException(status_code=500, detail=str(exc)) from exc

src/backend/base/langflow/api/v1/schemas.py

@@ -388,7 +388,6 @@ class ConfigResponse(BaseModel):
     public_flow_cleanup_interval: int
     public_flow_expiration: int
     event_delivery: Literal["polling", "streaming", "direct"]
-    voice_mode_enabled: bool
 
 
 class CancelFlowResponse(BaseModel):

src/backend/base/langflow/api/v1/voice_mode.py

@@ -7,14 +7,13 @@ import traceback
 import uuid
 from collections import defaultdict
 from datetime import datetime, timezone
-from functools import partial
+from functools import lru_cache, partial
 from typing import Any
 from uuid import UUID, uuid4
 
 import numpy as np
 import requests
 import sqlalchemy
-import webrtcvad
 import websockets
 from cryptography.fernet import InvalidToken
 from elevenlabs import ElevenLabs
@@ -83,6 +82,13 @@ CLIENT_TO_LF = "Client → LF"
 # --- Helper Functions ---
 
 
+@lru_cache(maxsize=1)
+def get_vad():
+    import webrtcvad
+
+    return webrtcvad.Vad(mode=3)
+
+
 async def authenticate_and_get_openai_key(session: DbSession, user: User, websocket: WebSocket):
     """Authenticate the user using a token or API key and retrieve the OpenAI API key.
 
@@ -129,7 +135,7 @@ class VoiceConfig:
         self.elevenlabs_model = "eleven_multilingual_v2"
         self.elevenlabs_client = None
         self.elevenlabs_key = None
-        self.barge_in_enabled = False
+        self._barge_in_enabled = False
         self.progress_enabled = True
 
         self.default_openai_realtime_session = {
@@ -151,6 +157,12 @@ class VoiceConfig:
         }
         self.openai_realtime_session: dict[str, Any] = {}
 
+    @property
+    def barge_in_enabled(self):
+        # Later on we may want to tie this value
+        # to the availability of the webrtcvad package.
+        return self._barge_in_enabled
+
     def get_session_dict(self):
         return dict(self.default_openai_realtime_session)
 
@@ -765,11 +777,11 @@ async def flow_as_tool_websocket(
             vad_queue: asyncio.Queue = asyncio.Queue()
             vad_audio_buffer = bytearray()
             bot_speaking_flag = [False]
-            vad = webrtcvad.Vad(mode=3)
 
             async def process_vad_audio() -> None:
                 nonlocal vad_audio_buffer
                 last_speech_time = datetime.now(tz=timezone.utc)
+                vad = get_vad()
                 while True:
                     base64_data = await vad_queue.get()
                     raw_chunk_24k = base64.b64decode(base64_data)

src/frontend/src/controllers/API/queries/config/use-get-config.ts

@@ -21,7 +21,6 @@ export interface ConfigResponse {
   webhook_polling_interval: number;
   serialization_max_items_length: number;
   event_delivery: EventDeliveryType;
-  voice_mode_enabled: boolean;
 }
 
 export const useGetConfig: useQueryFunctionType<undefined, ConfigResponse> = (
@@ -45,9 +44,7 @@ export const useGetConfig: useQueryFunctionType<undefined, ConfigResponse> = (
     (state) => state.setWebhookPollingInterval,
   );
   const setEventDelivery = useUtilityStore((state) => state.setEventDelivery);
-  const setVoiceModeEnabled = useFlowsManagerStore(
-    (state) => state.setVoiceModeEnabled,
-  );
+
   const { query } = UseRequestProcessor();
 
   const getConfigFn = async () => {
@@ -69,7 +66,6 @@ export const useGetConfig: useQueryFunctionType<undefined, ConfigResponse> = (
         data.webhook_polling_interval ?? DEFAULT_POLLING_INTERVAL,
       );
       setEventDelivery(data.event_delivery ?? EventDeliveryType.POLLING);
-      setVoiceModeEnabled(Boolean(data.voice_mode_enabled));
     }
     return data;
   };

src/frontend/src/modals/IOModal/components/chatView/chatInput/components/voice-assistant/components/voice-button.tsx

@@ -1,7 +1,6 @@
 import ForwardedIconComponent from "@/components/common/genericIconComponent";
 import { Button } from "@/components/ui/button";
 import { ICON_STROKE_WIDTH } from "@/constants/constants";
-import useFlowsManagerStore from "@/stores/flowsManagerStore";
 import { useVoiceStore } from "@/stores/voiceStore";
 
 interface VoiceButtonProps {
@@ -12,13 +11,6 @@ const VoiceButton = ({ toggleRecording }: VoiceButtonProps) => {
   const setNewSessionCloseVoiceAssistant = useVoiceStore(
     (state) => state.setNewSessionCloseVoiceAssistant,
   );
-  const voiceModeEnabled = useFlowsManagerStore(
-    (state) => state.voiceModeEnabled,
-  );
-
-  if (!voiceModeEnabled) {
-    return null;
-  }
 
   return (
     <>

src/frontend/src/stores/flowsManagerStore.ts

@@ -17,8 +17,6 @@ const past = {};
 const future = {};
 
 const useFlowsManagerStore = create<FlowsManagerStoreType>((set, get) => ({
-  voiceModeEnabled: false,
-  setVoiceModeEnabled: (voiceModeEnabled: boolean) => set({ voiceModeEnabled }),
   IOModalOpen: false,
   setIOModalOpen: (IOModalOpen: boolean) => {
     set({ IOModalOpen });

src/frontend/src/types/zustand/flowsManager/index.ts

@@ -29,8 +29,6 @@ export type FlowsManagerStoreType = {
   IOModalOpen: boolean;
   setIOModalOpen: (IOModalOpen: boolean) => void;
   resetStore: () => void;
-  voiceModeEnabled: boolean;
-  setVoiceModeEnabled: (voiceModeEnabled: boolean) => void;
 };
 
 export type UseUndoRedoOptions = {