如何使用Microsoft认知API实现语音转文字？及Node.js代码排障

Hey there! Let's tackle your two questions about Microsoft's Speech-to-Text API step by step.

First, let's cover the basics you'll need to get started, then walk through two common implementation approaches for Node.js:

Prerequisites #

• An Azure account (the free tier works great for testing)
• A Speech resource created in the Azure portal—this gives you an API key and region (like eastus or westeurope) for authentication
• Audio files that meet the API's requirements: WAV format with PCM encoding, 16kHz sample rate, and single channel (mono). These specs are the most reliable for consistent results.

Option 1: Use the Microsoft Speech SDK for Node.js #

This is the simplest and most robust method, as the SDK handles authentication, audio streaming, and error handling out of the box.

1. Install the SDK package:

   npm install microsoft-cognitiveservices-speech-sdk
   

2. Basic code example to transcribe a local WAV file:

   const sdk = require("microsoft-cognitiveservices-speech-sdk");
   const fs = require("fs");
   
   // Replace with your Speech resource key and region
   const speechConfig = sdk.SpeechConfig.fromSubscription("YOUR_API_KEY", "YOUR_REGION");
   speechConfig.speechRecognitionLanguage = "zh-CN"; // Set your target language
   
   // Point to your local WAV file
   const audioConfig = sdk.AudioConfig.fromWavFileInput(fs.readFileSync("./your-audio-file.wav"));
   
   // Initialize the recognizer
   const recognizer = new sdk.SpeechRecognizer(speechConfig, audioConfig);
   
   // Handle successful recognition
   recognizer.recognized = (s, e) => {
       if (e.result.reason === sdk.ResultReason.RecognizedSpeech) {
           console.log(`转文字结果: ${e.result.text}`);
       } else if (e.result.reason === sdk.ResultReason.NoMatch) {
           console.log(`未识别到有效语音: ${e.result.noMatchDetails}`);
       }
   };
   
   // Handle errors or cancellations
   recognizer.canceled = (s, e) => {
       console.log(`识别流程被取消: ${e.reason}`);
       if (e.reason === sdk.CancellationReason.Error) {
           console.log(`错误详情: ${e.errorDetails}`);
       }
       recognizer.stopContinuousRecognitionAsync();
   };
   
   // Start the recognition process
   recognizer.startContinuousRecognitionAsync();
   

Option 2: Call the REST API Directly #

If you prefer raw HTTP requests without the SDK, you can send the audio file directly to the Speech-to-Text endpoint. Here's a quick example using axios:

const axios = require("axios");
const fs = require("fs");

const apiKey = "YOUR_API_KEY";
const region = "YOUR_REGION";
const audioBuffer = fs.readFileSync("./your-audio-file.wav");

axios.post(
    `https://${region}.stt.speech.microsoft.com/speech/recognition/conversation/cognitiveservices/v1?language=zh-CN`,
    audioBuffer,
    {
        headers: {
            "Ocp-Apim-Subscription-Key": apiKey,
            "Content-Type": "audio/wav; codecs=audio/pcm; samplerate=16000",
        },
    }
)
.then(response => {
    console.log("转文字结果:", response.data.DisplayText);
})
.catch(error => {
    console.error("请求失败:", error.response?.data || error.message);
});

Since you didn't share the specific error code or your code snippet, let's go through the most common culprits:

• Authentication mistakes: Double-check that your API key and region match exactly what's listed in your Azure Speech resource. Typos here are the #1 cause of failures! Also, confirm the resource isn't disabled or expired.
• Audio format mismatch: The API is strict about specs. Use ffmpeg to convert your WAV to the required format if needed:

  ffmpeg -i input.wav -acodec pcm_s16le -ar 16000 -ac 1 output.wav
  

• File reading issues: Ensure your code points to the correct file path, and the Node.js process has read permissions for that file. Log the file size (e.g., console.log(fs.statSync("./file.wav").size)) to confirm it's loading properly.
• Incorrect headers (REST API only): If using the REST endpoint, make sure the Content-Type header includes the exact codec and samplerate (e.g., audio/wav; codecs=audio/pcm; samplerate=16000). Missing or wrong headers will trigger errors.
• Quota limits: Check your Azure portal to see if you've hit the free tier's request limit. If so, you may need to upgrade to a paid tier or wait for the quota to reset.
• Network restrictions: If you're behind a firewall or proxy, ensure it allows traffic to Azure Speech service endpoints (e.g., *.speech.microsoft.com).

If you can share your specific error message and a snippet of your code, I can help narrow down the issue even further!

内容的提问来源于stack exchange，提问作者omega