diff --git a/conf/config.json b/conf/config.json
index c59ec99..cad1a9e 100644
--- a/conf/config.json
+++ b/conf/config.json
@@ -3,6 +3,7 @@
 	"sample_rate":16000,
 	"remove_silence":false,
 	"modelname":"F5-TTS",
+	"device":"cuda:0",
 	"ref_audio_fn":"$[workdir]$/samples/test_zh_1_ref_short.wav",
 	"ref_text":"对，这就是我，万人敬仰的太乙真人。",
 	"cross_fade_duration":0
diff --git a/f5tts.py b/f5tts.py
index fb16a09..f961d45 100644
--- a/f5tts.py
+++ b/f5tts.py
@@ -1,4 +1,5 @@
-import time
+from time import time
+import torch
 from pathlib import Path
 import codecs
 import re
@@ -30,11 +31,6 @@ ode_method = "euler"
 sway_sampling_coef = -1.0
 speed = 1.0
 
-F5TTS_model_cfg = dict(
-	dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4
-)
-E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)
-
 def chunk_text(text, max_chars=135):
 	"""
 	Splits the input text into chunks, each with a maximum number of characters.
@@ -69,8 +65,9 @@ class F5TTS:
 		self.remove_silence = config.remove_silence
 		self.modelname = config.modelname
 		self.ref_audio_fn = config.ref_audio_fn
+		self.zmq_url = config.zmq_url
 		self.ref_text = config.ref_text
-		self.model= self.load_model(self.modelname)
+		self.device = config.device
 		self.cross_fade_duration = config.cross_fade_duration
 		self.gen_ref_audio()
 		self.gen_ref_text()
@@ -78,7 +75,7 @@ class F5TTS:
 		try:	
 			print(f"Load vocos from local path {vocos_local_path}")
 			vocos = Vocos.from_hparams(f"{vocos_local_path}/config.yaml")
-			state_dict = torch.load(f"{vocos_local_path}/pytorch_model.bin", map_location=device)
+			state_dict = torch.load(f"{vocos_local_path}/pytorch_model.bin", map_location=self.device)
 			vocos.load_state_dict(state_dict)
 			vocos.eval()
 			self.vocos = vocos
@@ -87,19 +84,25 @@ class F5TTS:
 			vocos = Vocos.from_pretrained("charactr/vocos-mel-24khz")
 			self.vocos = vocos
 
+        self.F5TTS_model_cfg = dict(
+            dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4
+        )
+        self.E2TTS_model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)
+		self.model= self.load_model(self.modelname)
+
 
 	def gen_ref_audio(self):
 		"""
 		gen ref_audio
 		"""
-		audio, sr = torchaudio.load(ref_audio)
+		audio, sr = torchaudio.load(self.ref_audio_fn)
 		if audio.shape[0] > 1:
 			audio = torch.mean(audio, dim=0, keepdim=True)
 		rms = torch.sqrt(torch.mean(torch.square(audio)))
 		if rms < target_rms:
 			audio = audio * target_rms / rms
-		if sr != target_sample_rate:
-			resampler = torchaudio.transforms.Resample(sr, target_sample_rate)
+		if sr != self.sample_rate:
+			resampler = torchaudio.transforms.Resample(sr, self.sample_rate)
 			audio = resampler(audio)
 		self.ref_audio = audio
 
@@ -138,29 +141,29 @@ class F5TTS:
 				method="euler"
 			),
 			vocab_char_map=vocab_char_map,
-		).to(device)
+		).to(self.device)
 
-		model = load_checkpoint(model, ckpt_path, device, use_ema = True)
+		model = load_checkpoint(model, ckpt_path, self.device, use_ema = True)
 
 		return model
 
 	def load_model(self, model):
 		if model == 'F5-TTS':
-			ret = self._load_model(model, "F5TTS_Base", DiT, F5TTS_model_cfg, 1200000)
+			ret = self._load_model(model, "F5TTS_Base", DiT, self.F5TTS_model_cfg, 1200000)
 			return ret
 		if model == 'E2-TTS':
-			return self._load_model(model, "E2TTS_Base", UNetT, E2TTS_model_cfg, 1200000)
+			return self._load_model(model, "E2TTS_Base", UNetT, self.E2TTS_model_cfg, 1200000)
 
 	def split_text(self, text):
 		max_chars = int(len(self.ref_text.encode('utf-8')) / (self.ref_audio.shape[-1] / self.sample_rate) * (25 - self.ref_audio.shape[-1] / self.sample_rate))
 		gen_text_batches = chunk_text(gen_text, max_chars=max_chars)
 		print('ref_text', ref_text)
 
-	def inference(self, prmpt, stream=False):
+	def inference(self, prompt, stream=False):
 		generated_waves = []
 		max_chars = int(len(self.ref_text.encode('utf-8')) / (self.ref_audio.shape[-1] / self.sample_rate) * (25 - self.ref_audio.shape[-1] / self.sample_rate))
 		gen_text_batches = chunk_text(prompt, max_chars=max_chars)
-		for gen_text in gen_text_Batches:
+		for gen_text in gen_text_batches:
 			# Prepare the text
 			text_list = [self.ref_text + gen_text]
 			final_text_list = convert_char_to_pinyin(text_list)
@@ -173,6 +176,7 @@ class F5TTS:
 			duration = ref_audio_len + int(ref_audio_len / ref_text_len * gen_text_len / speed)
 
 			# inference
+			print(f'{self.device=}....')
 			with torch.inference_mode():
 				generated, _ = self.model.sample(
 					cond=self.ref_audio,
@@ -194,6 +198,7 @@ class F5TTS:
 				generated_wave = generated_wave.squeeze().cpu().numpy()
 				generated_waves.append(generated_wave)
 		if stream:
+			print(f'here ........{stream}')
 			return
 		if self.cross_fade_duration <= 0:
 			# Simply concatenate
@@ -201,7 +206,9 @@ class F5TTS:
 		else:
 			final_wave = self.cross_fade_wave(generated_waves)
 		fn = self.write_wave(final_wave)
-		return fn
+		print(f'here ........{stream}, {fn=}')
+		yield fn
+		return
 	
 	def cross_fade_wave(self, waves):
 		final_wave = generated_waves[0]
@@ -249,20 +256,35 @@ class F5TTS:
 		data = DictObject(**json.loads(msg))
 		print(data)
 		t1 = time()
-		f = self.inference(data.prompt)
+		for wav in self.inference(data.prompt, stream=data.stream):
+			if data.stream:
+				d = {
+					"reqid":data.reqid,
+					"b64wave":b64str(wav),
+					"finish":False
+				}
+				self.replier.send(json.dumps(d))
+			else:
+				t2 = time()
+				d = {
+					"reqid":data.reqid,
+					"audio_file":wav,
+					"time_cost":t2 - t1
+				}
+				print(f'{d}')
+				return json.dumps(d)
 		t2 = time()
 		d = {
-			"audio_file":f,
-			"time_cost":t2 - t1
+			"reqid":data.reqid,
+			"time_cost":t2 - t1,
+			"finish":True
 		}
-		print(f'{d}')
 		return json.dumps(d)
 
 
 if __name__ == '__main__':
 	workdir = os.getcwd()
 	config = getConfig(workdir)
-	print(f'{config=}')
 	tts = F5TTS()
 	print('here')
 	tts.run()
diff --git a/samples/test.wav b/samples/test.wav
new file mode 100644
index 0000000..4027401
Binary files /dev/null and b/samples/test.wav differ
diff --git a/samples/test_en_1_ref_short.wav b/samples/test_en_1_ref_short.wav
new file mode 100644
index 0000000..3c593c3
Binary files /dev/null and b/samples/test_en_1_ref_short.wav differ
diff --git a/samples/test_zh_1_ref_short.wav b/samples/test_zh_1_ref_short.wav
new file mode 100644
index 0000000..8cc055e
Binary files /dev/null and b/samples/test_zh_1_ref_short.wav differ
diff --git a/zmq_client.py b/zmq_client.py
new file mode 100644
index 0000000..ed0179d
--- /dev/null
+++ b/zmq_client.py
@@ -0,0 +1,43 @@
+import json
+import os
+
+from appPublic.dictObject import DictObject
+from appPublic.zmq_reqrep import ZmqRequester
+from appPublic.jsonConfig import getConfig
+from appPublic.uniqueID import getID
+
+zmq_url = "tcp://127.0.0.1:9999"
+from time import time
+
+class ASRClient:
+    def __init__(self, zmq_url):
+        self.zmq_url = zmq_url
+        self.requester = ZmqRequester(self.zmq_url)
+
+    def generate(self, prompt):
+        d = {
+            "prompt":prompt,
+            "reqid":getID()
+        }
+        msg = json.dumps(d)
+        resp = self.requester.send(msg)
+        if resp != None:
+            ret = json.loads(resp)
+            print(f'response={ret}')
+        else:
+            print(f'response is None')
+
+    def run(self):
+        print(f'running {self.zmq_url}')
+        while True:
+            print('input audio_file:')
+            af = input()
+            if len(af) > 0:
+                self.generate(af)
+        print('ended ...')
+
+if __name__ == '__main__':
+    workdir = os.getcwd()
+    config = getConfig(workdir)
+    asr = ASRClient(config.zmq_url or zmq_url)
+    asr.run()