Update app.py

app.py

@@ -13,7 +13,7 @@ from fastrtc import AsyncStreamHandler, Stream, wait_for_item
 from pydantic import BaseModel
 import uvicorn
 
-# --- Import get_space to detect Hugging Face Spaces (optional) ---
+# --- Import get_space (optional) ---
 from gradio.utils import get_space
 
 # --- Document processing and RAG libraries ---
@@ -37,7 +37,6 @@ current_dir = pathlib.Path(__file__).parent
 # 1. Document Ingestion & RAG Pipeline Setup
 # ====================================================
 
-# Folder containing PDFs, Word docs, and text files (place this folder alongside app.py)
 DOCS_FOLDER = current_dir / "docs"
 
 def extract_text_from_pdf(file_path: pathlib.Path) -> str:
@@ -78,20 +77,17 @@ def split_text(text: str, max_length: int = 500, overlap: int = 100) -> List[str
         start += max_length - overlap
     return chunks
 
-# Load and process documents
 documents = load_documents(DOCS_FOLDER)
 all_chunks = []
 for doc in documents:
     all_chunks.extend(split_text(doc))
 
-# Compute embeddings and build FAISS index
 embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
 chunk_embeddings = embedding_model.encode(all_chunks)
 embedding_dim = chunk_embeddings.shape[1]
 index = faiss.IndexFlatL2(embedding_dim)
 index.add(np.array(chunk_embeddings))
 
-# Setup a text-generation pipeline (using GPT-2 here as an example)
 generator = pipeline("text-generation", model="gpt2", max_length=256)
 
 def retrieve_context(query: str, k: int = 5) -> List[str]:
@@ -119,7 +115,6 @@ def generate_answer(query: str) -> str:
 stt_model = whisper.load_model("base", device="cpu")
 
 def speech_to_text(audio_array: np.ndarray, sample_rate: int = 16000) -> str:
-    # Convert int16 PCM to float32 normalized to [-1, 1]
     audio_float = audio_array.astype(np.float32) / 32768.0
     result = stt_model.transcribe(audio_float, fp16=False)
     return result["text"]
@@ -197,9 +192,17 @@ class RAGVoiceHandler(AsyncStreamHandler):
 # 4. Voice Streaming Setup & FastAPI Endpoints
 # ====================================================
 
-# For ZeroGPU spaces, supply a dummy RTC configuration.
-# (This avoids calling get_twilio_turn_credentials() which depends on NVML.)
-rtc_config = {"iceServers": [{"urls": "stun:stun.l.google.com:19302"}]}
+# Supply a dummy (but valid) RTC configuration to satisfy fastrtc.
+rtc_config = {
+    "iceServers": [
+        {"urls": "stun:stun.l.google.com:19302"},
+        {
+            "urls": "turn:turn.anyfirewall.com:443?transport=tcp",
+            "username": "webrtc",
+            "credential": "webrtc"
+        }
+    ]
+}
 
 stream = Stream(
     modality="audio",