Update app.py

app.py

@@ -1,29 +1,48 @@
 import os
 import io
 import asyncio
+import numpy as np
+import librosa
 import soundfile as sf
 import gradio as gr
 from google import genai
 
 client = genai.Client(api_key=os.getenv("GEMINI_API_KEY"))
 
-MODEL = "gemini-2.5-flash-native-audio-preview-09-2025"
+MODEL = "gemini-2.5-flash-native-audio-preview-12-2025"
 
 config = {
-    "response_modalities": ["AUDIO"]
+    "response_modalities": ["AUDIO"],
+    "system_instruction": "You are a helpful assistant and answer in a friendly tone.",
 }
 
-async def generate_audio_response(text: str):
+async def generate_audio_response(audio_bytes: bytes):
     async with client.aio.live.connect(model=MODEL, config=config) as session:
-        await session.send_client_content(
-            turns={"role": "user", "parts": [{"text": text}]},
-            turn_complete=True,
+        await session.send_realtime_input(
+            audio={"data": audio_bytes, "mime_type": "audio/pcm"}
         )
 
         audio_chunks = []
-        async for response in session.receive():
-            if response.data is not None:
-                audio_chunks.append(response.data)
+        last_receive_time = asyncio.get_event_loop().time()
+
+        while True:
+            turn = session.receive()
+            has_new = False
+            async for response in turn:
+                if response.server_content and response.server_content.model_turn:
+                    for part in response.server_content.model_turn.parts:
+                        if hasattr(part, "inline_data") and part.inline_data.data:
+                            audio_chunks.append(part.inline_data.data)
+                            has_new = True
+                            last_receive_time = asyncio.get_event_loop().time()
+
+            if audio_chunks and not has_new and (asyncio.get_event_loop().time() - last_receive_time > 3):
+                break
+
+            await asyncio.sleep(0.2)
+
+            if asyncio.get_event_loop().time() - last_receive_time > 30:
+                break
 
         full_audio = b''.join(audio_chunks)
         if not full_audio:
@@ -33,40 +52,51 @@ async def generate_audio_response(text: str):
         y, sr = sf.read(buf, channels=1, samplerate=24000, format="RAW", subtype="PCM_16", dtype="float32")
         return sr, y
 
-def process_text(input_text: str | None):
-    if not input_text or not input_text.strip():
-        return None, "Please enter a message."
+def process_audio(input_path: str | None):
+    if input_path is None:
+        return None, "Please upload a WAV file."
 
     try:
-        sr, audio_data = asyncio.run(generate_audio_response(input_text.strip()))
-        return (sr, audio_data), "Response generated successfully!"
+        y, orig_sr = librosa.load(input_path, sr=None, mono=True)
+        y = librosa.resample(y, orig_sr=orig_sr, target_sr=16000)
+        y_int = np.int16(y * 32767)
+        audio_bytes = y_int.tobytes()
+
+        sr, response_audio = asyncio.run(generate_audio_response(audio_bytes))
+
+        return (sr, response_audio), "Response generated successfully!"
     except Exception as e:
         return None, f"Error: {str(e)}"
 
 with gr.Blocks() as demo:
-    gr.Markdown("# Gemini Text to Spoken Audio")
-    gr.Markdown("Enter text → Gemini responds with audio")
-
-    input_text = gr.Textbox(
-        label="Your message",
-        placeholder="Hello? Gemini are you there?",
-        lines=3
-    )
+    gr.Markdown("# Gemini Live Audio-to-Audio Demo")
+    gr.Markdown("Upload a WAV file (spoken query). Gemini will respond with spoken audio.")
+
+    with gr.Row():
+        input_audio = gr.Audio(
+            label="Upload your query (WAV file)",
+            type="filepath",
+            sources=["upload"],
+            format="wav"
+        )
 
-    output_audio = gr.Audio(
-        label="Gemini spoken response",
-        type="numpy",
-        autoplay=True
-    )
+    with gr.Row():
+        output_audio = gr.Audio(
+            label="Gemini spoken response",
+            type="numpy",
+            autoplay=True
+        )
 
     status = gr.Textbox(label="Status")
 
     btn = gr.Button("Generate Response")
 
     btn.click(
-        fn=process_text,
-        inputs=input_text,
+        fn=process_audio,
+        inputs=input_audio,
         outputs=[output_audio, status]
     )
 
+    gr.Markdown("Example test file: https://storage.googleapis.com/generativeai-downloads/data/16000.wav")
+
 demo.launch()