Node.js SDK

Prerequisites

• Install Node.js 14 or later and npm, and configure the environment.

• Obtain a project Appkey and an access token. For project configuration, see Manage projects. For token acquisition, see Obtain an access token.

• Review the audio requirements and service parameters in the Real-time speech recognition API.

Install the SDK

Install the SDK in your project directory:

npm install alibabacloud-nls

The complete example uses CommonJS. In an ES module project, use a named import:

import { SpeechTranscription } from "alibabacloud-nls";

Code example

This example uses alibabacloud-nls 1.0.4 to recognize a local audio file once. Before running the example, set NLS_APP_KEY and NLS_TOKEN in the runtime environment to your project Appkey and access token.

Download the official SDK's sample audio test1.pcm to the program's working directory. The file contains Chinese numbers in 16 kHz, mono, 16-bit PCM format. For other audio, adjust the file path, audio parameters, and sending interval.

"use strict";

const fs = require("fs/promises");
const Nls = require("alibabacloud-nls");
const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));

async function main() {
  const appkey = process.env.NLS_APP_KEY;
  const token = process.env.NLS_TOKEN;
  if (!appkey || !token) {
    throw new Error("Set NLS_APP_KEY and NLS_TOKEN.");
  }
  const audio = await fs.readFile("test1.pcm");
  const client = new Nls.SpeechTranscription({
    url: "wss://nls-gateway.cn-shanghai.aliyuncs.com/ws/v1",
    appkey,
    token
  });
  let timer;
  let finished = false;

  try {
    await new Promise((resolve, reject) => {
      timer = setTimeout(() => reject(new Error("Recognition timed out.")), 60000);
      client.on("failed", (message) => reject(new Error(message)));
      client.on("closed", () => {
        if (!finished) reject(new Error("Connection closed before completion."));
      });
      client.on("changed", (message) => {
        console.log("Intermediate:", JSON.parse(message).payload.result);
      });
      client.on("end", (message) => {
        console.log("Sentence:", JSON.parse(message).payload.result);
      });
      client.on("completed", (message) => {
        finished = true;
        console.log("Transcription completed.");
        resolve(message);
      });

      (async () => {
        await client.start(client.defaultStartParams(), true, 6000);
        for (let offset = 0; offset < audio.length; offset += 1024) {
          if (!client.sendAudio(audio.subarray(offset, offset + 1024))) {
            throw new Error("Audio could not be sent.");
          }
          await sleep(32);
        }
        await client.close();
      })().catch(reject);
    });
  } finally {
    clearTimeout(timer);
    client.shutdown();
  }
}

main().catch((error) => {
  console.error(error.message || error);
  process.exitCode = 1;
});

The example sends 1,024 bytes every 32 milliseconds to match the PCM data rate and limits the call to 60 seconds. The end event provides the final result for each sentence. The completed event indicates that the entire task has ended.

Interfaces and parameters

const client = new Nls.SpeechTranscription({ url, appkey, token });

{
  "format": "pcm",
  "sample_rate": 16000,
  "enable_intermediate_result": true,
  "enable_punctuation_prediction": true,
  "enable_inverse_text_normalization": true
}