From 7e0652f8d5701e5952a1c81770de4e06e0019f9b Mon Sep 17 00:00:00 2001
From: 游雁 <zhifu.gzf@alibaba-inc.com>
Date: 星期四, 27 四月 2023 10:30:13 +0800
Subject: [PATCH] websocket
---
funasr/runtime/python/websocket/ASR_client.py | 38 +++-
funasr/runtime/python/websocket/ASR_server_streaming.py | 261 ++++++++++++++++++++++++++++++++
funasr/runtime/python/websocket/ASR_server_streaming_asr.py | 149 ++++++++++++++++++
3 files changed, 438 insertions(+), 10 deletions(-)
diff --git a/funasr/runtime/python/websocket/ASR_client.py b/funasr/runtime/python/websocket/ASR_client.py
index cc0e7b6..b0abfc7 100644
--- a/funasr/runtime/python/websocket/ASR_client.py
+++ b/funasr/runtime/python/websocket/ASR_client.py
@@ -1,5 +1,4 @@
-
-# import websocket #鍖哄埆鏈嶅姟绔繖閲屾槸 websocket-client搴�
+# -*- encoding: utf-8 -*-
import time
import websockets
import asyncio
@@ -50,18 +49,21 @@
rate=RATE,
input=True,
frames_per_buffer=CHUNK)
-
+ is_speaking = True
while True:
data = stream.read(CHUNK)
+ data = data.decode('ISO-8859-1')
+ message = json.dumps({"chunk": args.chunk_size, "is_speaking": is_speaking, "audio": data})
- voices.put(data)
+ voices.put(message)
#print(voices.qsize())
await asyncio.sleep(0.01)
# 鍏朵粬鍑芥暟鍙互閫氳繃璋冪敤send(data)鏉ュ彂閫佹暟鎹紝渚嬪锛�
async def record_from_scp():
+ import wave
global voices
if args.audio_in.endswith(".scp"):
f_scp = open(args.audio_in)
@@ -71,15 +73,31 @@
for wav in wavs:
wav_splits = wav.strip().split()
wav_path = wav_splits[1] if len(wav_splits) > 1 else wav_splits[0]
- bytes = open(wav_path, "rb")
- bytes = bytes.read()
-
+ # bytes_f = open(wav_path, "rb")
+ # bytes_data = bytes_f.read()
+ with wave.open(wav_path, "rb") as wav_file:
+ # 鑾峰彇闊抽鍙傛暟
+ params = wav_file.getparams()
+ # 鑾峰彇澶翠俊鎭殑闀垮害
+ # header_length = wav_file.getheaders()[0][1]
+ # 璇诲彇闊抽甯ф暟鎹紝璺宠繃澶翠俊鎭�
+ # wav_file.setpos(header_length)
+ frames = wav_file.readframes(wav_file.getnframes())
+
+ # 灏嗛煶棰戝抚鏁版嵁杞崲涓哄瓧鑺傜被鍨嬬殑鏁版嵁
+ audio_bytes = bytes(frames)
stride = int(args.chunk_size/1000*16000*2)
- chunk_num = (len(bytes)-1)//stride + 1
+ chunk_num = (len(audio_bytes)-1)//stride + 1
+ print(stride)
+ is_speaking = True
for i in range(chunk_num):
+ if i == chunk_num-1:
+ is_speaking = False
beg = i*stride
- data_chunk = bytes[beg:beg+stride]
- voices.put(data_chunk)
+ data = audio_bytes[beg:beg+stride]
+ data = data.decode('ISO-8859-1')
+ message = json.dumps({"chunk": args.chunk_size, "is_speaking": is_speaking, "audio": data})
+ voices.put(message)
# print("data_chunk: ", len(data_chunk))
# print(voices.qsize())
diff --git a/funasr/runtime/python/websocket/ASR_server_streaming.py b/funasr/runtime/python/websocket/ASR_server_streaming.py
new file mode 100644
index 0000000..b7c54f7
--- /dev/null
+++ b/funasr/runtime/python/websocket/ASR_server_streaming.py
@@ -0,0 +1,261 @@
+import asyncio
+import json
+import websockets
+import time
+from queue import Queue
+import threading
+import argparse
+
+from modelscope.pipelines import pipeline
+from modelscope.utils.constant import Tasks
+from modelscope.utils.logger import get_logger
+import logging
+import tracemalloc
+import numpy as np
+
+tracemalloc.start()
+
+logger = get_logger(log_level=logging.CRITICAL)
+logger.setLevel(logging.CRITICAL)
+
+
+websocket_users = set() #缁存姢瀹㈡埛绔垪琛�
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--host",
+ type=str,
+ default="0.0.0.0",
+ required=False,
+ help="host ip, localhost, 0.0.0.0")
+parser.add_argument("--port",
+ type=int,
+ default=10095,
+ required=False,
+ help="grpc server port")
+parser.add_argument("--asr_model",
+ type=str,
+ default="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch",
+ help="model from modelscope")
+parser.add_argument("--vad_model",
+ type=str,
+ default="damo/speech_fsmn_vad_zh-cn-16k-common-pytorch",
+ help="model from modelscope")
+
+parser.add_argument("--punc_model",
+ type=str,
+ default="damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
+ help="model from modelscope")
+parser.add_argument("--ngpu",
+ type=int,
+ default=1,
+ help="0 for cpu, 1 for gpu")
+
+args = parser.parse_args()
+
+print("model loading")
+
+def load_bytes(input):
+ middle_data = np.frombuffer(input, dtype=np.int16)
+ middle_data = np.asarray(middle_data)
+ if middle_data.dtype.kind not in 'iu':
+ raise TypeError("'middle_data' must be an array of integers")
+ dtype = np.dtype('float32')
+ if dtype.kind != 'f':
+ raise TypeError("'dtype' must be a floating point type")
+
+ i = np.iinfo(middle_data.dtype)
+ abs_max = 2 ** (i.bits - 1)
+ offset = i.min + abs_max
+ array = np.frombuffer((middle_data.astype(dtype) - offset) / abs_max, dtype=np.float32)
+ return array
+
+# vad
+inference_pipeline_vad = pipeline(
+ task=Tasks.voice_activity_detection,
+ model=args.vad_model,
+ model_revision=None,
+ output_dir=None,
+ batch_size=1,
+ mode='online',
+ ngpu=args.ngpu,
+)
+# param_dict_vad = {'in_cache': dict(), "is_final": False}
+
+# # asr
+# param_dict_asr = {}
+# # param_dict["hotword"] = "灏忎簲 灏忎簲鏈�" # 璁剧疆鐑瘝锛岀敤绌烘牸闅斿紑
+# inference_pipeline_asr = pipeline(
+# task=Tasks.auto_speech_recognition,
+# model=args.asr_model,
+# param_dict=param_dict_asr,
+# ngpu=args.ngpu,
+# )
+# if args.punc_model != "":
+# # param_dict_punc = {'cache': list()}
+# inference_pipeline_punc = pipeline(
+# task=Tasks.punctuation,
+# model=args.punc_model,
+# model_revision=None,
+# ngpu=args.ngpu,
+# )
+# else:
+# inference_pipeline_punc = None
+
+
+inference_pipeline_asr_online = pipeline(
+ task=Tasks.auto_speech_recognition,
+ model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
+ model_revision=None)
+
+
+print("model loaded")
+
+
+
+async def ws_serve(websocket, path):
+ #speek = Queue()
+ frames = [] # 瀛樺偍鎵�鏈夌殑甯ф暟鎹�
+ frames_online = [] # 瀛樺偍鎵�鏈夌殑甯ф暟鎹�
+ buffer = [] # 瀛樺偍缂撳瓨涓殑甯ф暟鎹紙鏈�澶氫袱涓墖娈碉級
+ RECORD_NUM = 0
+ global websocket_users
+ speech_start, speech_end = False, False
+ # 璋冪敤asr鍑芥暟
+ websocket.param_dict_vad = {'in_cache': dict(), "is_final": False}
+ websocket.param_dict_punc = {'cache': list()}
+ websocket.speek = Queue() #websocket 娣诲姞杩涢槦鍒楀璞� 璁゛sr璇诲彇璇煶鏁版嵁鍖�
+ websocket.send_msg = Queue() #websocket 娣诲姞涓槦鍒楀璞� 璁﹚s鍙戦�佹秷鎭埌瀹㈡埛绔�
+ websocket_users.add(websocket)
+ # ss = threading.Thread(target=asr, args=(websocket,))
+ # ss.start()
+
+ websocket.param_dict_asr_online = {"cache": dict(), "is_final": False}
+ websocket.speek_online = Queue() # websocket 娣诲姞杩涢槦鍒楀璞� 璁゛sr璇诲彇璇煶鏁版嵁鍖�
+ ss_online = threading.Thread(target=asr_online, args=(websocket,))
+ ss_online.start()
+
+ try:
+ async for data in websocket:
+ #voices.put(message)
+ #print("put")
+ #await websocket.send("123")
+
+ data = json.loads(data)
+ # message = data["data"]
+ message = bytes(data['audio'], 'ISO-8859-1')
+ chunk = data["chunk"]
+ chunk_num = 600//chunk
+ is_speaking = data["is_speaking"]
+ websocket.param_dict_vad["is_final"] = not is_speaking
+ buffer.append(message)
+ if len(buffer) > 2:
+ buffer.pop(0) # 濡傛灉缂撳瓨瓒呰繃涓や釜鐗囨锛屽垯鍒犻櫎鏈�鏃╃殑涓�涓�
+
+ if speech_start:
+ # frames.append(message)
+ frames_online.append(message)
+ # RECORD_NUM += 1
+ if len(frames_online) % chunk_num == 0:
+ audio_in = b"".join(frames_online)
+ websocket.speek_online.put(audio_in)
+ frames_online = []
+
+ speech_start_i, speech_end_i = vad(message, websocket)
+ #print(speech_start_i, speech_end_i)
+ if speech_start_i:
+ # RECORD_NUM += 1
+ speech_start = speech_start_i
+ # frames = []
+ # frames.extend(buffer) # 鎶婁箣鍓�2涓闊虫暟鎹揩鍔犲叆
+ frames_online = []
+ # frames_online.append(message)
+ frames_online.extend(buffer)
+ # RECORD_NUM += 1
+ websocket.param_dict_asr_online["is_final"] = False
+ if speech_end_i:
+ speech_start = False
+ # audio_in = b"".join(frames)
+ # websocket.speek.put(audio_in)
+ # frames = [] # 娓呯┖鎵�鏈夌殑甯ф暟鎹�
+ frames_online = []
+ websocket.param_dict_asr_online["is_final"] = True
+ # buffer = [] # 娓呯┖缂撳瓨涓殑甯ф暟鎹紙鏈�澶氫袱涓墖娈碉級
+ # RECORD_NUM = 0
+ if not websocket.send_msg.empty():
+ await websocket.send(websocket.send_msg.get())
+ websocket.send_msg.task_done()
+
+
+ except websockets.ConnectionClosed:
+ print("ConnectionClosed...", websocket_users) # 閾炬帴鏂紑
+ websocket_users.remove(websocket)
+ except websockets.InvalidState:
+ print("InvalidState...") # 鏃犳晥鐘舵��
+ except Exception as e:
+ print("Exception:", e)
+
+
+# def asr(websocket): # ASR鎺ㄧ悊
+# global inference_pipeline_asr
+# # global param_dict_punc
+# global websocket_users
+# while websocket in websocket_users:
+# if not websocket.speek.empty():
+# audio_in = websocket.speek.get()
+# websocket.speek.task_done()
+# if len(audio_in) > 0:
+# rec_result = inference_pipeline_asr(audio_in=audio_in)
+# if inference_pipeline_punc is not None and 'text' in rec_result:
+# rec_result = inference_pipeline_punc(text_in=rec_result['text'], param_dict=websocket.param_dict_punc)
+# # print(rec_result)
+# if "text" in rec_result:
+# message = json.dumps({"mode": "offline", "text": rec_result["text"]})
+# websocket.send_msg.put(message) # 瀛樺叆鍙戦�侀槦鍒� 鐩存帴璋冪敤send鍙戦�佷笉浜�
+#
+# time.sleep(0.1)
+
+
+def asr_online(websocket): # ASR鎺ㄧ悊
+ global inference_pipeline_asr_online
+ # global param_dict_punc
+ global websocket_users
+ while websocket in websocket_users:
+ if not websocket.speek_online.empty():
+ audio_in = websocket.speek_online.get()
+ websocket.speek_online.task_done()
+ if len(audio_in) > 0:
+ # print(len(audio_in))
+ audio_in = load_bytes(audio_in)
+ # print(audio_in.shape)
+ rec_result = inference_pipeline_asr_online(audio_in=audio_in, param_dict=websocket.param_dict_asr_online)
+
+ # print(rec_result)
+ if "text" in rec_result:
+ if rec_result["text"] != "sil" and rec_result["text"] != "waiting_for_more_voice":
+ message = json.dumps({"mode": "online", "text": rec_result["text"]})
+ websocket.send_msg.put(message) # 瀛樺叆鍙戦�侀槦鍒� 鐩存帴璋冪敤send鍙戦�佷笉浜�
+
+ time.sleep(0.1)
+
+def vad(data, websocket): # VAD鎺ㄧ悊
+ global inference_pipeline_vad, param_dict_vad
+ #print(type(data))
+ # print(param_dict_vad)
+ segments_result = inference_pipeline_vad(audio_in=data, param_dict=websocket.param_dict_vad)
+ # print(segments_result)
+ # print(param_dict_vad)
+ speech_start = False
+ speech_end = False
+
+ if len(segments_result) == 0 or len(segments_result["text"]) > 1:
+ return speech_start, speech_end
+ if segments_result["text"][0][0] != -1:
+ speech_start = True
+ if segments_result["text"][0][1] != -1:
+ speech_end = True
+ return speech_start, speech_end
+
+
+start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None)
+asyncio.get_event_loop().run_until_complete(start_server)
+asyncio.get_event_loop().run_forever()
\ No newline at end of file
diff --git a/funasr/runtime/python/websocket/ASR_server_streaming_asr.py b/funasr/runtime/python/websocket/ASR_server_streaming_asr.py
new file mode 100644
index 0000000..396597e
--- /dev/null
+++ b/funasr/runtime/python/websocket/ASR_server_streaming_asr.py
@@ -0,0 +1,149 @@
+import asyncio
+import json
+import websockets
+import time
+from queue import Queue
+import threading
+import argparse
+
+from modelscope.pipelines import pipeline
+from modelscope.utils.constant import Tasks
+from modelscope.utils.logger import get_logger
+import logging
+import tracemalloc
+import numpy as np
+
+tracemalloc.start()
+
+logger = get_logger(log_level=logging.CRITICAL)
+logger.setLevel(logging.CRITICAL)
+
+
+websocket_users = set() #缁存姢瀹㈡埛绔垪琛�
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--host",
+ type=str,
+ default="0.0.0.0",
+ required=False,
+ help="host ip, localhost, 0.0.0.0")
+parser.add_argument("--port",
+ type=int,
+ default=10095,
+ required=False,
+ help="grpc server port")
+parser.add_argument("--asr_model",
+ type=str,
+ default="damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-pytorch",
+ help="model from modelscope")
+parser.add_argument("--vad_model",
+ type=str,
+ default="damo/speech_fsmn_vad_zh-cn-16k-common-pytorch",
+ help="model from modelscope")
+
+parser.add_argument("--punc_model",
+ type=str,
+ default="damo/punc_ct-transformer_zh-cn-common-vad_realtime-vocab272727",
+ help="model from modelscope")
+parser.add_argument("--ngpu",
+ type=int,
+ default=1,
+ help="0 for cpu, 1 for gpu")
+
+args = parser.parse_args()
+
+print("model loading")
+
+def load_bytes(input):
+ middle_data = np.frombuffer(input, dtype=np.int16)
+ middle_data = np.asarray(middle_data)
+ if middle_data.dtype.kind not in 'iu':
+ raise TypeError("'middle_data' must be an array of integers")
+ dtype = np.dtype('float32')
+ if dtype.kind != 'f':
+ raise TypeError("'dtype' must be a floating point type")
+
+ i = np.iinfo(middle_data.dtype)
+ abs_max = 2 ** (i.bits - 1)
+ offset = i.min + abs_max
+ array = np.frombuffer((middle_data.astype(dtype) - offset) / abs_max, dtype=np.float32)
+ return array
+
+inference_pipeline_asr_online = pipeline(
+ task=Tasks.auto_speech_recognition,
+ # model='damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online',
+ model='damo/speech_paraformer_asr_nat-zh-cn-16k-common-vocab8404-online',
+ model_revision=None)
+
+
+print("model loaded")
+
+
+
+async def ws_serve(websocket, path):
+ frames_online = []
+ global websocket_users
+ websocket.send_msg = Queue()
+ websocket_users.add(websocket)
+ websocket.param_dict_asr_online = {"cache": dict()}
+ websocket.speek_online = Queue()
+ ss_online = threading.Thread(target=asr_online, args=(websocket,))
+ ss_online.start()
+ try:
+ async for message in websocket:
+ message = json.loads(message)
+ audio = bytes(message['audio'], 'ISO-8859-1')
+ chunk = message["chunk"]
+ chunk_num = 500//chunk
+ is_speaking = message["is_speaking"]
+ websocket.param_dict_asr_online["is_final"] = not is_speaking
+ frames_online.append(audio)
+
+ if len(frames_online) % chunk_num == 0 or not is_speaking:
+ audio_in = b"".join(frames_online)
+ websocket.speek_online.put(audio_in)
+ frames_online = []
+
+ if not websocket.send_msg.empty():
+ await websocket.send(websocket.send_msg.get())
+ websocket.send_msg.task_done()
+
+
+ except websockets.ConnectionClosed:
+ print("ConnectionClosed...", websocket_users) # 閾炬帴鏂紑
+ websocket_users.remove(websocket)
+ except websockets.InvalidState:
+ print("InvalidState...") # 鏃犳晥鐘舵��
+ except Exception as e:
+ print("Exception:", e)
+
+
+
+def asr_online(websocket): # ASR鎺ㄧ悊
+ global inference_pipeline_asr_online
+ global websocket_users
+ while websocket in websocket_users:
+ if not websocket.speek_online.empty():
+ audio_in = websocket.speek_online.get()
+ websocket.speek_online.task_done()
+ if len(audio_in) > 0:
+ # print(len(audio_in))
+ audio_in = load_bytes(audio_in)
+ # print(audio_in.shape)
+ print(websocket.param_dict_asr_online["is_final"])
+ rec_result = inference_pipeline_asr_online(audio_in=audio_in, param_dict=websocket.param_dict_asr_online)
+ if websocket.param_dict_asr_online["is_final"]:
+ websocket.param_dict_asr_online["cache"] = dict()
+
+ print(rec_result)
+ if "text" in rec_result:
+ if rec_result["text"] != "sil" and rec_result["text"] != "waiting_for_more_voice":
+ message = json.dumps({"mode": "online", "text": rec_result["text"]})
+ websocket.send_msg.put(message) # 瀛樺叆鍙戦�侀槦鍒� 鐩存帴璋冪敤send鍙戦�佷笉浜�
+
+ time.sleep(0.005)
+
+
+start_server = websockets.serve(ws_serve, args.host, args.port, subprotocols=["binary"], ping_interval=None)
+asyncio.get_event_loop().run_until_complete(start_server)
+asyncio.get_event_loop().run_forever()
\ No newline at end of file
--
Gitblit v1.9.1