Azure Cognitive Services语音转写异常：生成的TXT文件为空

尝试将音频文件转写内容保存至TXT文件，但生成的文件为空。控制台仅输出RECOGNIZING: hello，未触发recognized回调写入文件，实现代码如下：

speech_config = speechsdk.SpeechConfig(subscription=speech_key, region=service_region)
audio_config = speechsdk.audio.AudioConfig(filename=audio_file)

speech_recognizer = speechsdk.SpeechRecognizer(speech_config=speech_config, audio_config=audio_config)
done = False

# Define a callback function to handle the result
def handle_result(result):
    print("Got result:", result.text)
    with open("transcript.txt", "a") as f:
        f.write(result.text + "\n")

# Start the continuous transcription process
speech_recognizer.recognizing.connect(lambda evt: print('RECOGNIZING: {}'.format(evt.result.text)))
speech_recognizer.recognized.connect(handle_result)
speech_recognizer.session_stopped.connect(lambda evt: print('SESSION STOPPED {}'.format(evt)))

speech_recognizer.start_continuous_recognition()
time.sleep(15) 
speech_recognizer.stop_continuous_recognition()

解决步骤 #

• 替换固定sleep为事件驱动等待
  固定time.sleep(15)会强制终止未完成的识别过程，导致recognized回调无法触发。改用会话结束事件控制停止时机：

  done = False
  
  def stop_recognition(evt):
      print('SESSION STOPPED {}'.format(evt))
      speech_recognizer.stop_continuous_recognition()
      nonlocal done
      done = True
  
  # 绑定会话停止和取消事件
  speech_recognizer.session_stopped.connect(stop_recognition)
  speech_recognizer.canceled.connect(stop_recognition)
  
  speech_recognizer.start_continuous_recognition()
  # 等待识别完成
  while not done:
      time.sleep(0.5)
  

• 添加错误捕获回调排查问题
  未监听canceled回调会错过识别过程中的错误信息，添加后可快速定位授权、音频格式或网络问题：

  def handle_canceled(evt):
      print('CANCELED: {}'.format(evt.reason))
      if evt.reason == speechsdk.CancellationReason.Error:
          print('ERROR details: {}'.format(evt.error_details))
  
  speech_recognizer.canceled.connect(handle_canceled)
  

• 确保音频触发最终识别结果
  recognized回调对应稳定的最终识别结果，若音频过短或无结尾静音，SDK可能无法确认识别完成。可以给音频末尾添加2-3秒静音，或检查音频文件是否完整。

• 验证音频格式兼容性
  确认音频文件为Speech SDK支持的格式（如WAV、MP3），若格式不兼容会导致识别中断。建议转换为16kHz、16位、单声道的PCM WAV格式测试。

内容的提问来源于stack exchange，提问作者Genaro Raúl Mateu Lagunes