Google Speech-to-Text single_utterance模式超时设置问题求助

方法一：使用API原生的speech_start_timeout参数 #

你之前混淆了参数作用，speech_end_timeout是语音结束后等待后续输入的超时，而**speech_start_timeout才是专门设置等待语音开始的超时时间**。配合enable_voice_activity_events启用语音活动检测，就能在指定时间内未检测到语音时，让API主动结束流式请求，避免程序卡死。

配置示例：

from google.cloud import speech_v1p1beta1 as speech

client = speech.SpeechClient()
recog_config = speech.RecognitionConfig(
    encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
    sample_rate_hertz=16000,
    language_code="zh-CN",
    single_utterance=True,
)

streaming_config = speech.StreamingRecognitionConfig(
    config=recog_config,
    interim_results=False,
    enable_voice_activity_events=True,
    speech_start_timeout=30.0  # 30秒内未检测到语音则触发超时
)

超时触发时，API会返回包含语音活动事件的响应或抛出对应RPC错误，你可以捕获该情况结束循环。

方法二：客户端本地定时器兜底 #

如果API原生参数不生效，可在启动流式识别的同时，启动本地定时器，超时后主动取消流式请求。

代码示例（Python）：

import threading
from google.cloud import speech_v1p1beta1 as speech

def handle_transcription():
    client = speech.SpeechClient()
    recog_config = speech.RecognitionConfig(
        encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
        sample_rate_hertz=16000,
        language_code="zh-CN",
        single_utterance=True,
    )
    streaming_config = speech.StreamingRecognitionConfig(
        config=recog_config,
        interim_results=False
    )

    # 超时回调：主动取消流式请求
    def timeout_cancel(stream):
        print("30秒无语音输入，结束识别")
        stream.cancel()

    # 替换为你的实际麦克风音频流生成逻辑
    audio_generator = your_microphone_audio_generator()
    requests = (speech.StreamingRecognizeRequest(audio_content=chunk) for chunk in audio_generator)

    responses = client.streaming_recognize(streaming_config, requests)
    # 启动30秒定时器
    timer = threading.Timer(30.0, timeout_cancel, args=[responses])
    timer.start()

    try:
        for response in responses:
            # 收到识别结果，立即取消定时器
            timer.cancel()
            for result in response.results:
                print(f"转写结果：{result.alternatives[0].transcript}")
    except Exception as e:
        print(f"识别结束：{str(e)}")
    finally:
        # 确保定时器被取消，避免内存泄漏
        timer.cancel()

方法三：捕获RPC超时异常 #

流式识别过程中，若API因超时断开连接，会抛出gRPC的DEADLINE_EXCEEDED异常，捕获该异常即可结束循环，避免程序卡死。

示例代码：

from grpc import RpcError, StatusCode

try:
    for response in responses:
        for result in response.results:
            print(f"转写结果：{result.alternatives[0].transcript}")
except RpcError as e:
    if e.code() == StatusCode.DEADLINE_EXCEEDED:
        print("无语音输入，超时退出")
    else:
        # 其他异常正常抛出
        raise e

内容的提问来源于stack exchange，提问作者Antonio Bono