Python3 使用 websockets 调用阿里云实时语音识别（qbit）

头像
qbit
     重庆
    阅读 11 分钟

    前言

    • 技术栈

      Python  3.11.8
websockets  15.0.1
aliyun-python-sdk-core  2.16.0
nls  1.0.0
    • 截至 2025.3.13，nls.NlsSpeechTranscriber 不支持异步调用
    • 使用 asyncio.run 或 loop.create_task 将异步调用转化为同步调用

    • 后文中，为保持字节流和字符串的一致性，定义了不同格式的结束符

      b'$END$' client -> server
 '$END$' server -> client

    材料准备

    • 从 GitHub 下载 nls 目录和 tests 目录里面的 test1.pcm 文件
      image.png
    • 将 nls 目录放到 site-packages 目录下，相当于安装 nls 库

    示例代码

    • server 端代码（sync 同步版本）

      # encoding: utf-8
# author: qbit
# date: 2025-03-13
# summary: websocket 服务端结束语言识别请求，调用阿里云返回识别的文字结果

import time
import json
import nls
from websockets.sync.server import serve

URL = "wss://nls-gateway-cn-shanghai.aliyuncs.com/ws/v1"
TOKEN = "your_token"  # 参考 https://help.aliyun.com/document_detail/450255.html 获取token
APPKEY = "your_key"  # 获取Appkey请前往控制台：https://nls-portal.console.aliyun.com/applist

class FuncCallback:
  r"""nls.NlsSpeechTranscriber 需要的回调函数"""

  def __init__(self, ):
      self.result = ""
      self.state = ""

  def on_sentence_begin(self, message, *args):
      print("on_sentence_begin")

  def on_sentence_end(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "sentence_end"
      print(f"on_sentence_end: {self.result}")
      websocket = args[0]
      websocket.send(self.result)

  def on_start(self, message, *args):
      print("on_start")

  def on_error(self, message, *args):
      print(f"on_error from aliyun {message}")
      self.state = "error"

  def on_close(self, *args):
      print("on_close")

  def on_result_chg(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "result_chg"
      print(f"on_result_chg: {self.result}")
      websocket = args[0]
      websocket.send(self.result)

  def on_completed(self, message, *args):
      print("on_completed")
      self.state = "completed"

def asr(websocket):
  r"""语言识别服务，Automatic Speech Recognition (ASR)"""
  print(
      f"Client connected: {websocket.remote_address[0]}:{websocket.remote_address[1]}"
  )
  callback = FuncCallback()
  sr = nls.NlsSpeechTranscriber(
      url=URL,
      token=TOKEN,
      appkey=APPKEY,
      on_sentence_begin=callback.on_sentence_begin,
      on_sentence_end=callback.on_sentence_end,
      on_start=callback.on_start,
      on_result_changed=callback.on_result_chg,
      on_completed=callback.on_completed,
      on_error=callback.on_error,
      on_close=callback.on_close,
      callback_args=[websocket],
  )
  sr.start(
      aformat="pcm",              # 支持格式 pcm, opu, opus
      enable_intermediate_result=True,
      enable_punctuation_prediction=True,
      enable_inverse_text_normalization=True,
  )
  for message in websocket:
      if message == b"$END$":
          break
      sr.send_audio(message)

  sr.stop()

  while callback.state not in ('completed', 'error'):
      time.sleep(0.1)
      print(f"callback.state: {callback.state}")
  websocket.send("$END$")        # 发送自定义结束标志

def main():
  host = "127.0.0.1"
  port = 8765
  with serve(asr, host, port) as server:
      print(f"Server started {host}:{port}...")
      server.serve_forever()

if __name__ == "__main__":
  main()

    • client 端代码

      # encoding: utf-8
# author: qbit
# date: 2025-03-13
# summary: websocket 客户端模拟实时语音请求

import threading
from websockets.sync.client import connect

def send(websocket):
  r"""
  测试语言识别
  Automatic Speech Recognition (ASR)
  """
  with open("./audio/test1.pcm", "rb") as f:
      data = f.read()
  __slices = zip(*(iter(data),) * 6400)
  for i in __slices:
      websocket.send(bytes(i))
  websocket.send(b"$END$")    # 发送自定义结束标志

def recv(websocket):
  r"""
  测试语言识别
  Automatic Speech Recognition (ASR)
  """
  while True:
      message = websocket.recv()
      print(f"Received message: {message}")
      if message == "$END$":
          break

if __name__ == "__main__":
  uri = "ws://localhost:8765"
  print(f"Connecting to {uri}...")
  with connect(uri) as websocket:
      t1 = threading.Thread(target=send, args=(websocket,))
      t2 = threading.Thread(target=recv, args=(websocket,))
      t1.start()
      t2.start()
      t1.join()
      t2.join()

    运行与输出

    • 先启动 server 代码，再启动 client 代码

    • server 输出

      Server started 127.0.0.1:8765...
Client connected: 127.0.0.1:53387
on_start
on_sentence_begin
on_result_chg: 因
on_result_chg: 一
on_result_chg: 12
on_result_chg: 123
on_result_chg: 1234
on_result_chg: 12345
on_result_chg: 123456
on_result_chg: 1234567
on_result_chg: 12345678
on_result_chg: 1 2 3 4 5 6 7 8 9 10
on_sentence_end: 1 2 3 4 5 6 7 8 9 10。
on_completed
on_close

    • client 输出

      Connecting to ws://localhost:8765...
Received message: 因
Received message: 一
Received message: 12
Received message: 123
Received message: 1234
Received message: 12345
Received message: 123456
Received message: 1234567
Received message: 12345678
Received message: 1 2 3 4 5 6 7 8 9 10
Received message: 1 2 3 4 5 6 7 8 9 10。
Received message: $END$

    async 异步版 Server 代码

    • 本节的异步版本代码，可以与 fastapi.WebSocket 结合使用

    • asyncio.run 版本

      # encoding: utf-8
# author: qbit
# date: 2025-03-13
# summary: websocket 服务端结束语言识别请求，调用阿里云返回识别的文字结果

import json
import nls
import asyncio
from websockets.asyncio.server import serve

URL = "wss://nls-gateway-cn-shanghai.aliyuncs.com/ws/v1"
TOKEN = "your_token"  # 参考 https://help.aliyun.com/document_detail/450255.html 获取token
APPKEY = "your_key"  # 获取Appkey请前往控制台：https://nls-portal.console.aliyun.com/applist

class FuncCallback:
  r"""nls.NlsSpeechTranscriber 需要的回调函数"""
  def __init__(self, ):
      self.result = ""
      self.state = ""

  def on_sentence_begin(self, message, *args):
      print("on_sentence_begin")

  def on_sentence_end(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "sentence_end"
      print(f"on_sentence_end: {self.result}")
      websocket = args[0]
      asyncio.run(websocket.send(self.result))    # 将异步调用转化为同步调用

  def on_start(self, message, *args):
      print("on_start")

  def on_error(self, message, *args):
      print(f"on_error from aliyun {message}")
      self.state = "error"

  def on_close(self, *args):
      print("on_close")

  def on_result_chg(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "result_chg"
      print(f"on_result_chg: {self.result}")
      websocket = args[0]
      asyncio.run(websocket.send(self.result))    # 将异步调用转化为同步调用

  def on_completed(self, message, *args):
      print("on_completed")
      self.state = "completed"

async def asr(websocket):
  callback = FuncCallback()
  sr = nls.NlsSpeechTranscriber(
      url=URL,
      token=TOKEN,
      appkey=APPKEY,
      on_sentence_begin=callback.on_sentence_begin,
      on_sentence_end=callback.on_sentence_end,
      on_start=callback.on_start,
      on_result_changed=callback.on_result_chg,
      on_completed=callback.on_completed,
      on_error=callback.on_error,
      on_close=callback.on_close,
      callback_args=[websocket],
  )
  sr.start(
      aformat="pcm",
      enable_intermediate_result=True,
      enable_punctuation_prediction=True,
      enable_inverse_text_normalization=True,
  )
  async for message in websocket:
      if message == b"$END$":
          break
      sr.send_audio(message)

  sr.stop()

  while callback.state not in ("completed", "error"):
      await asyncio.sleep(0.1)
      print(f"callback.state: {callback.state}")

  await websocket.send("$END$")  # 发送自定义结束标志

async def main():
  host = "127.0.0.1"
  port = 8765
  async with serve(asr, host, port) as server:
      print(f"Server started {host}:{port}...")
      await server.serve_forever()

if __name__ == "__main__":
  asyncio.run(main())

    • 因为 asyncio.run 新开了事件循环有些重，花了不少时间弄出下面的 loop.create_task 版本

      # encoding: utf-8
# author: qbit
# date: 2025-03-14
# summary: websocket 服务端结束语言识别请求，调用阿里云返回识别的文字结果

import json
import nls
import asyncio
from websockets.asyncio.server import serve

URL = "wss://nls-gateway-cn-shanghai.aliyuncs.com/ws/v1"
TOKEN = "your_token"  # 参考 https://help.aliyun.com/document_detail/450255.html 获取token
APPKEY = "your_key"  # 获取Appkey请前往控制台：https://nls-portal.console.aliyun.com/applist

class FuncCallback:
  r"""nls.NlsSpeechTranscriber 需要的回调函数"""

  def __init__(self):
      self.result = ""
      self.state = ""

  def on_sentence_begin(self, message, *args):
      print("on_sentence_begin")

  def on_sentence_end(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "sentence_end"
      print(f"on_sentence_end: {self.result}")
      websocket = args[0]
      loop: asyncio.windows_events.ProactorEventLoop = args[1]
      loop.create_task(websocket.send(self.result))

  def on_start(self, message, *args):
      print("on_start")

  def on_error(self, message, *args):
      print(f"on_error from aliyun {message}")
      self.state = "error"

  def on_close(self, *args):
      print("on_close")

  def on_result_chg(self, message, *args):
      self.result = json.loads(message)["payload"]["result"]
      self.state = "result_chg"
      print(f"on_result_chg: {self.result}")
      websocket = args[0]
      loop: asyncio.windows_events.ProactorEventLoop = args[1]
      loop.create_task(websocket.send(self.result))

  def on_completed(self, message, *args):
      print("on_completed")
      self.state = "completed"

async def asr(websocket):
  callback = FuncCallback()
  # loop = asyncio.get_running_loop()
  loop = asyncio.get_event_loop()
  sr = nls.NlsSpeechTranscriber(
      url=URL,
      token=TOKEN,
      appkey=APPKEY,
      on_sentence_begin=callback.on_sentence_begin,
      on_sentence_end=callback.on_sentence_end,
      on_start=callback.on_start,
      on_result_changed=callback.on_result_chg,
      on_completed=callback.on_completed,
      on_error=callback.on_error,
      on_close=callback.on_close,
      callback_args=[websocket, loop],
  )
  sr.start(
      aformat="pcm",  # 支持格式 pcm, opu, opus
      enable_intermediate_result=True,
      enable_punctuation_prediction=True,
      enable_inverse_text_normalization=True,
  )

  async for message in websocket:
      if message == b"$END$":
          break
      sr.send_audio(message)

  sr.stop()
  # 下一行代码很重要！没有的话 FuncCallback 里面的 loop.create_task 不会被触发
  await asyncio.sleep(0)

  while callback.state not in ("completed", "error"):
      await asyncio.sleep(0.1)
      print(f"callback.state: {callback.state}")

  await websocket.send("$END$")  # 发送自定义结束标志

async def main():
  host = "127.0.0.1"
  port = 8765
  async with serve(asr, host, port) as server:
      print(f"Server started {host}:{port}...")
      await server.serve_forever()

if __name__ == "__main__":
  asyncio.run(main())

    相关阅读

    本文出自 qbit sanp
    websocketasync
    qbit
    271 声望279 粉丝
    « 上一篇
    安装 apk 和 obb 文件（qbit）

    引用和评论

    推荐阅读
    头像
    百万级群聊的设计实践

    vivo互联网技术赞 1阅读 1.1k

    头像
    Nginx转发WebSocket到Spring Cloud Gateway并路由到Spring Boot服务

    Pursuer丶阅读 3.3k评论 2

    头像
    SpringBoot微信扫码登录（小程序版）

    Pursuer丶阅读 2.8k

    头像
    以太坊与比特币：2023年的挑战与机遇

    健身的油条_ADoT6阅读 554

    头像
    websocket webworker教程及应用

    阿芯爱编程阅读 550

    头像
    股票websocket提供股票实时数据的推送功能

    CryptoRzz阅读 410

    0 条评论
    得票最新