使用GCP构建语音转文本遇metadata获取错误，求解决方案

问题描述 #

使用GCP构建语音转文本功能时，持续触发错误：

Getting metadata from plugin failed with error: key must be a string, a buffer or an object

错误提示该故障并非临时问题，且未主动配置重试逻辑。

现有代码（speech.ts） #

import { SpeechClient } from '@google-cloud/speech';
import axios from 'axios';

export async function transcribeAudioFromUrl(
  audioUrl: string,
): Promise<string> {
  try {
    // Configure the Speech-to-Text client with your credentials
    const speechClient = new SpeechClient({
      keyFilename:
        'config/client_secret_206787547924-3a46gf902osirh0h7fm5ud7l6e41mtap.apps.googleusercontent.com.json', // Replace with the correct path to your credentials
    });

    // Convert the binary audio data into a buffer
    const audioBuffer = Buffer.from(audioUrl);

    console.log({ audioBuffer });

    // Perform audio transcription
    const [responseSpeech] = await speechClient.recognize({
      audio: {
        content: audioBuffer,
      },
    });

    const transcription = responseSpeech.results
      .map((result) => result.alternatives[0].transcript)
      .join('\n');

    return transcription;
  } catch (error: any) {
    console.log(error);
  }
}

错误详情 #

Error: 2 UNKNOWN: Getting metadata from plugin failed with error: key must be a string, a buffer or an object
    at callErrorFromStatus (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/call.ts:81:17)
    at Object.onReceiveStatus (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/client.ts:356:55)
    at Object.onReceiveStatus (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/client-interceptors.ts:454:34)
    at Object.onReceiveStatus (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/client-interceptors.ts:416:48)
    at /Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/resolving-call.ts:111:24
    at processTicksAndRejections (node:internal/process/task_queues:77:11)
for call at
    at ServiceClientImpl.makeUnaryRequest (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/client.ts:326:42)
    at ServiceClientImpl.<anonymous> (/Users/duca/html/sofia/node_modules/@grpc/grpc-js/src/make-client.ts:189:15)
    at /Users/duca/html/sofia/node_modules/@google-cloud/speech/build/src/v1/speech_client.js:229:29
    at /Users/duca/html/sofia/node_modules/google-gax/build/src/normalCalls/timeout.js:44:16
    at repeat (/Users/duca/html/sofia/node_modules/google-gax/build/src/normalCalls/retries.js:80:25)
    at /Users/duca/html/sofia/node_modules/google-gax/build/src/normalCalls/retries.js:118:13
    at OngoingCallPromise.call (/Users/duca/html/sofia/node_modules/google-gax/build/src/call.js:67:27)
    at NormalApiCaller.call (/Users/duca/html/sofia/node_modules/google-gax/build/src/normalCalls/normalApiCaller.js:34:19)
    at /Users/duca/html/sofia/node_modules/google-gax/build/src/createApiCall.js:84:30
    at processTicksAndRejections (node:internal/process/task_queues:95:5) {
  code: 2,
  details: 'Getting metadata from plugin failed with error: key must be a string, a buffer or an object',
  metadata: Metadata { internalRepr: Map(0) {}, options: {} },
  note: 'Exception occurred in retry method that was not classified as transient'
}

问题分析与修复方案 #

核心问题 #

1. 凭证类型错误：使用的client_secret_xxx.json是OAuth客户端凭证，GCP Speech-to-Text需要的是服务账号JSON密钥文件，二者格式不兼容导致身份验证失败。
2. 音频处理逻辑错误：直接将音频URL转换为Buffer，实际得到的是URL字符串的二进制编码，而非音频文件的二进制数据。
3. 缺少识别配置参数：recognize请求未指定音频编码、采样率等核心配置，API无法正确解析音频。

修复步骤 #

1. 替换为正确的服务账号密钥

• 登录GCP控制台，进入「IAM与管理」→「服务账号」
• 创建或选择拥有Cloud Speech-to-Text Admin权限的服务账号
• 点击「添加密钥」→「创建新密钥」，选择JSON格式下载，替换代码中keyFilename的路径

2. 修复音频下载逻辑

通过axios下载音频的二进制数据，再转换为Buffer。

3. 添加语音识别配置参数

根据音频实际属性（编码格式、采样率、语言）配置config字段。

修复后完整代码 #

import { SpeechClient } from '@google-cloud/speech';
import axios from 'axios';

export async function transcribeAudioFromUrl(
  audioUrl: string,
): Promise<string> {
  try {
    // 使用服务账号密钥文件（注意：不是OAuth客户端凭证）
    const speechClient = new SpeechClient({
      keyFilename: 'config/service-account-key.json', // 替换为你的服务账号密钥路径
    });

    // 下载音频文件的二进制数据
    const response = await axios.get(audioUrl, { responseType: 'arraybuffer' });
    const audioBuffer = Buffer.from(response.data);

    // 配置语音识别参数（根据你的音频实际情况调整）
    const config = {
      encoding: 'MP3', // 音频编码格式，如MP3、LINEAR16（对应WAV）等
      sampleRateHertz: 16000, // 音频采样率
      languageCode: 'zh-CN', // 识别语言，如'en-US'、'zh-CN'
    };

    const request = {
      audio: {
        content: audioBuffer,
      },
      config: config,
    };

    // 执行语音识别
    const [responseSpeech] = await speechClient.recognize(request);

    const transcription = responseSpeech.results
      .map((result) => result.alternatives[0].transcript)
      .join('\n');

    return transcription;
  } catch (error: any) {
    console.error('转录失败:', error);
    throw error; // 抛出错误供调用方处理，避免静默失败
  }
}

额外注意事项 #

• 确保服务账号拥有Cloud Speech-to-Text Admin或Cloud Speech-to-Text Editor权限
• 音频参数（编码、采样率）必须与实际音频一致，否则识别会失败或结果不准确
• 检查依赖版本，建议使用最新版的@google-cloud/speech和axios

内容的提问来源于stack exchange，提问作者Duca