diff --git a/rtcllm/a2a.py b/rtcllm/a2a.py
new file mode 100644
index 0000000..0315f31
--- /dev/null
+++ b/rtcllm/a2a.py
@@ -0,0 +1,41 @@
+import io
+from mini_omni.inference import OmniInference
+from av import AudioFrame
+
+class LLMAudioStreamTrack(AudioStreamTrack):
+	def __init__(self, ckpt_dir):
+		super().__init__()
+		self.ckpt_dir = ckpt_dir
+		self.oi = OmniInference(ckpt_dir=chpt_dir)
+		self.audio_iters = []
+		self.cur_iters = None
+	
+	async def recv(self):
+		b = self.get_audio_bytes()
+		if b is None:
+			return b
+		frame = AudioFrame.from_ndarray(io.BytesIO(b), format='s16', layout='mono')
+		return frame
+
+	def set_cur_audio_iter(self):
+		if len(self.audio_iters) == 0:
+			return False
+		self.cur_iters = self.audio_iters[0]
+		self.audio_iters.remove(self.cur_iters)
+		rteturn True
+
+	def get_audio_bytes(self):
+			if self.cur_iters is None:
+				if not set_cur_audio_iter():
+					return None
+			try:
+				b = next(self.cur_iters)
+				return b
+			except StopIteration:
+				self.cur_iters = None
+				return self.get_audio_bytes()
+
+	def _feed(self, audio_file):
+		x = self.oi.run_AT_batch_stream(audio_file)
+		self.audio_iters.append(x)
+	
diff --git a/rtcllm/rtc.py b/rtcllm/rtc.py
index 1333f8f..5c8fccb 100644
--- a/rtcllm/rtc.py
+++ b/rtcllm/rtc.py
@@ -16,7 +16,7 @@ from websockets.client import connect
 import websockets
 from stt import asr
 from vad import AudioTrackVad
-from aav import MyAudioTrack, MyVideoTrack
+from a2a import LLMAudioStreamTrack
 
 videos = ['./1.mp4', './2.mp4']
 
@@ -45,6 +45,8 @@ class RTCLLM:
 	def __init__(self, ws_url, iceServers):
 		# self.stt_model = WhisperModel('large-v3', device="cuda", compute_type="float16")
 		self.ws_url = ws_url
+		self.llmtrack = LLMAudioStreamTrack('/d/models/mini-omni')
+		self.feed = awaitify(self.llmtrack._feed)
 		self.iceServers = iceServers
 		self.peers = DictObject()
 		self.vid = 0
@@ -100,9 +102,7 @@ class RTCLLM:
 		self.onlineList = data.onlineList
 	
 	async def vad_voiceend(self, peer, audio):
-		ret = await asr(audio)
-		# await peer.dc.send(txt)
-	
+		ret = await self.feed(audio)
 
 	async def auto_accept_call(self, data):
 		opts = DictObject(iceServers=self.iceServers)
@@ -111,6 +111,7 @@ class RTCLLM:
 			'info':data['from'],
 			'pc':pc
 		})
+		pc.addTrack(self.llmtrack)
 		await self.ws_send(json.dumps({'type':'callAccepted', 'to':data['from']}))
 	
 	async def pc_track(self, peerid, track):
@@ -122,14 +123,6 @@ class RTCLLM:
 			peer.vadtrack = vadtrack
 			vadtrack.start_vad()
 
-	def play_random_media(self, vt, at):
-		i = random.randint(0,1)
-		player = MediaPlayer(videos[i])
-		vt.set_source(player)
-		at.set_source(player)
-		f = partial(self.play_random_media, vt, at)
-		self.loop.call_later(180, f)
-
 	async def pc_connectionState_changed(self, peerid):
 		peer = self.peers[peerid]
 		pc = peer.pc
@@ -159,11 +152,6 @@ class RTCLLM:
 		pc.on("connectionstatechange", partial(self.pc_connectionState_changed, data['from'].id))
 		pc.on('track', partial(self.pc_track, data['from'].id))
 		pc.on('icecandidate', partial(self.on_icecandidate, pc, data['from']))
-		peer.audiotrack = MyAudioTrack()
-		peer.videotrack = MyVideoTrack()
-		pc.addTrack(peer.audiotrack)
-		pc.addTrack(peer.videotrack)
-		self.play_random_media(peer.videotrack, peer.audiotrack)
 
 		offer = RTCSessionDescription(** data.offer)
 		await pc.setRemoteDescription(offer)