Create app.py

app.py

@@ -0,0 +1,95 @@
+import gradio as gr
+from faster_whisper import WhisperModel
+
+# ---- Load Whisper once (choose "tiny" for speed; "base" for better accuracy) ----
+ASR_MODEL_SIZE = "tiny"
+asr_model = WhisperModel(ASR_MODEL_SIZE, device="cpu", compute_type="int8")
+
+def transcribe_audio(audio_path: str) -> str:
+    if not audio_path:
+        return ""
+    segments, _info = asr_model.transcribe(audio_path, vad_filter=True)
+    text = " ".join(seg.text.strip() for seg in segments).strip()
+    return text
+
+# ---- Your concierge brain (v1 = rule/KB; later connect OpenAI/RAG) ----
+def generate_reply(user_text: str) -> str:
+    t = (user_text or "").lower()
+    if "restaurant" in t or "dinner" in t or "menu" in t:
+        return (
+            "Restaurant info:\n"
+            "• Dinner: 7:00 PM – 10:30 PM\n"
+            "• You can view today’s dinner menu here (demo): Veg options, grills, noodles, desserts.\n"
+            "Would you like vegetarian items or something spicy?"
+        )
+    if "bar" in t or "close" in t:
+        return "The bar closes at 11:00 PM (demo). Would you like the last-order time as well?"
+    if "towel" in t:
+        return "Sure — I can place a request for extra towels. How many towels do you need?"
+    return "I can help with timings, directions, menus, and service requests. What would you like to know?"
+
+def handle_concierge(mode: str, text_input: str, audio_input: str):
+    if mode == "Voice":
+        user_text = transcribe_audio(audio_input)
+        if not user_text:
+            return "", "❌ I couldn't hear anything. Please try again.", ""
+        reply = generate_reply(user_text)
+        return user_text, reply, reply
+    else:
+        user_text = (text_input or "").strip()
+        if not user_text:
+            return "", "❌ Please type a question.", ""
+        reply = generate_reply(user_text)
+        return user_text, reply, reply
+
+with gr.Blocks(title="AI Digital Concierge (Text + Voice)") as demo:
+    gr.Markdown("# 🏨 AI Digital Concierge (Prototype)\nText or Voice — ask about facilities, menus, timings, and requests.")
+
+    mode = gr.Radio(["Text", "Voice"], value="Text", label="Choose interaction mode")
+
+    # Inputs
+    text_in = gr.Textbox(label="Type your question (Text mode)", placeholder="e.g., Can I see the dinner menu?")
+    audio_in = gr.Audio(label="Speak your question (Voice mode)", sources=["microphone", "upload"], type="filepath")
+
+    ask_btn = gr.Button("Ask Concierge", variant="primary")
+
+    # Outputs
+    transcribed = gr.Textbox(label="Transcribed text (Voice mode)", interactive=False)
+    response = gr.Textbox(label="Concierge Response", lines=6, interactive=False, elem_id="agent_response")
+    response_state = gr.State("")  # holds latest response for speak button
+
+    # Browser-side TTS (no server TTS)
+    gr.HTML("""
+    <div style="margin-top: 8px;">
+      <button
+        style="padding:10px 14px; border-radius:10px; border:1px solid #ccc; cursor:pointer;"
+        onclick="
+          const t = document.getElementById('agent_response')?.value || '';
+          if (!t) { alert('No response to read yet.'); return; }
+          window.speechSynthesis.cancel();
+          const u = new SpeechSynthesisUtterance(t);
+          u.rate = 1.0; u.pitch = 1.0; u.lang = 'en-US';
+          window.speechSynthesis.speak(u);
+        "
+      >
+        🔊 Speak Response
+      </button>
+      <button
+        style="padding:10px 14px; border-radius:10px; border:1px solid #ccc; cursor:pointer; margin-left:8px;"
+        onclick="window.speechSynthesis.cancel();"
+      >
+        ⏹ Stop
+      </button>
+      <div style="font-size:12px; color:#666; margin-top:6px;">
+        Voice output uses your browser’s speech engine (fast, no server load).
+      </div>
+    </div>
+    """)
+
+    ask_btn.click(
+        handle_concierge,
+        inputs=[mode, text_in, audio_in],
+        outputs=[transcribed, response, response_state],
+    )
+
+demo.launch()