Azure语音转文本无法识别PC麦克风录制的WAV音频问题

1. 排查PC录制WAV的格式兼容性 #

Azure语音转文本服务对输入音频有严格格式要求，默认支持：

• 采样率：16kHz
• 位深：16位
• 声道：单声道（mono）

PC麦克风录制的WAV通常默认是44.1kHz/立体声，这类不符合要求的格式会导致服务无法识别音频内容，进而出现无响应的情况。

• 验证方法：用Audacity等音频工具打开录制的WAV，查看格式参数。
• 修复方法：将音频转换为16kHz/16位/单声道的WAV格式后再传入脚本。

2. 完善代码的错误与状态监听 #

你的代码缺失**canceled事件监听**，这是排查无响应问题的核心——如果音频格式不合法或出现其他服务端错误，Azure会触发canceled事件，但当前代码未处理该事件，导致Promise永远无法resolve，脚本陷入停滞。

同时建议添加超时兜底逻辑，防止因异常情况导致无限等待。修改后的代码需补充以下内容：

// 新增canceled事件监听
speechRecognizer.canceled = (s, e) => {
  console.log(`CANCELED: Reason=${e.reason}`);
  if (e.reason === sdk.CancellationReason.Error) {
    console.log(`CANCELED: ErrorCode=${e.errorCode}`);
    console.log(`CANCELED: ErrorDetails=${e.errorDetails}`);
    reject(new Error(`Recognition canceled: ${e.errorDetails}`));
  }
  speechRecognizer.stopContinuousRecognitionAsync();
};

// 新增超时兜底（5分钟，可按需调整）
setTimeout(() => {
  if (!response) {
    speechRecognizer.stopContinuousRecognitionAsync();
    reject(new Error("Recognition timed out"));
  }
}, 300000);

3. 优化音频读取方式 #

使用fs.readFileSync读取大音频文件可能引发内存占用过高的问题，建议改用流读取方式提升稳定性：

const audioConfig = sdk.AudioConfig.fromWavFileInput(fs.createReadStream(input_file));

完整修改后的代码示例 #

const fs = require('fs');
const sdk = require("microsoft-cognitiveservices-speech-sdk");

async function transcribe_speech(input_file) {
  const speechConfig = sdk.SpeechConfig.fromSubscription(process.env.SPEECH_KEY, process.env.SPEECH_REGION);
  speechConfig.speechRecognitionLanguage = "he-IL";
  // 改用流读取音频文件
  const audioConfig = sdk.AudioConfig.fromWavFileInput(fs.createReadStream(input_file));
  const speechRecognizer = new sdk.SpeechRecognizer(speechConfig, audioConfig);
  
  return new Promise((resolve, reject) => {
    let response = "";
    console.log("In promise")
    
    // 新增canceled事件监听
    speechRecognizer.canceled = (s, e) => {
      console.log(`CANCELED: Reason=${e.reason}`);
      if (e.reason === sdk.CancellationReason.Error) {
        console.log(`CANCELED: ErrorCode=${e.errorCode}`);
        console.log(`CANCELED: ErrorDetails=${e.errorDetails}`);
        reject(new Error(`Recognition canceled: ${e.errorDetails}`));
      }
      speechRecognizer.stopContinuousRecognitionAsync();
    };

    speechRecognizer.recognized = (s, e) => {
      console.log("In speech recognizer")
      if (e.result.reason === sdk.ResultReason.RecognizedSpeech) {
        response += e.result.text;
        console.log("In recognized speech")
      } else if (e.result.reason === sdk.ResultReason.NoMatch) {
        console.log("NOMATCH: Speech could not be recognized.");
      }
    };

    speechRecognizer.sessionStopped = (s, e) => {
      console.log("Session stopped.");
      speechRecognizer.stopContinuousRecognitionAsync();
      resolve(response);
    };

    speechRecognizer.startContinuousRecognitionAsync();
    
    // 超时兜底逻辑
    setTimeout(() => {
      if (!response) {
        speechRecognizer.stopContinuousRecognitionAsync();
        reject(new Error("Recognition timed out after 5 minutes"));
      }
    }, 300000);
  });
}

module.exports = { transcribe_speech }

transcribe_speech("./uploads/recorded_from_pc.wav")
.then((msg) => {
    console.log("Recognition completed. Final response:");
    console.log(msg);
  })
  .catch((err) => {
    console.log("Recognition error:");
    console.error(err);
});

内容的提问来源于stack exchange，提问作者Martin Chapman