讯飞语音听写WebApi(流式)【React Native版】

假设已有 Base64 编码的音频文件(16kHz, s16le, pcm)

1、获取websocket url

import * as CryptoJS from 'crypto-js';

/**
 * 获取websocket url
 */
const getWebSocketUrl = () => {
	const config = {
		// 请求地址
		hostUrl: "wss://iat-api.xfyun.cn/v2/iat",
		host: "iat-api.xfyun.cn",
		//在控制台-我的应用-语音听写（流式版）获取
		appid: '',
		apiSecret: '',  // 替换为你的 APISecret
		apiKey: '',		// 替换为你的 APIKey
		file: "./16k_10.pcm", //请填写您的音频文件路径
		uri: "/v2/iat",
		highWaterMark: 1280
	}
	const date = new Date().toUTCString()
	// 鉴权签名
	let signatureOrigin = `host: ${config.host}\ndate: ${date}\nGET ${config.uri} HTTP/1.1`
	let signatureSha = CryptoJS.HmacSHA256(signatureOrigin, config.apiSecret);
	let signature = CryptoJS.enc.Base64.stringify(signatureSha)
	let authorizationOrigin = `api_key="${config.apiKey}", algorithm="hmac-sha256", headers="host date request-line", signature="${signature}"`
	let authStr = CryptoJS.enc.Base64.stringify(CryptoJS.enc.Utf8.parse(authorizationOrigin))
	let wssUrl = `${config.hostUrl}?authorization=${authStr}&date=${encodeURIComponent(date).replace(/%2C/g, ',').replace(/%3A/g, ':')}&host=${config.host}`

	return wssUrl
}

2、建立连接并发送数据

需要注意的是如果 base64 数据很大不能一次发送完成，就需要将 base64 重新转为二进制数据后按需要的大小切割，然后将切割后的二进制片段再重新转为 base64 编码后发送，否则服务端无法正确读取数据。

import { Buffer } from 'buffer'; // 导入 Buffer

// 如果有socket地址
const url = getWebSocketUrl();
const ws = new WebSocket(url);
ws.onopen = () => {
	console.log("🚀 ~ 连接成功")
	const params = {
		common: { app_id: '' },//appId
		business: {
			language: "zh_cn",
			domain: "iat",
			accent: "mandarin",
			vad_eos: 5000,
			dwa: "wpgs",
		},
		data: {
			status: 0,
			format: "audio/L16;rate=16000",
			encoding: "raw",
		},
	};
	ws.send(JSON.stringify(WebApiSocketStore.params));

	try {
		// 解码 Base64 字符串为二进制数据
		const audioPcmData = Buffer.from(data.base64Pcm, 'base64');

		let offset = 0;
		while (offset < audioPcmData.length) {
			const subString = audioPcmData.slice(offset, offset + 1280)
			offset += 1280
			const isEnd = offset >= audioPcmData.length;

			const btoa = Buffer.from(subString, 'utf-8').toString('base64');

			const dataAudio = JSON.stringify({
				data: {
					status: isEnd ? 2 : 1,
					format: "audio/L16;rate=16000",
					encoding: "raw",
					audio: `${btoa}`, // 将二进制字符串转换为 Base64
				},
			})
			ws?.send(dataAudio);
		}
	} catch (error) {
		console.log('🚀 ~ Error reading file:', error);
	}
};
ws.onmessage = (e) => {
	// a message was received
	console.log("🚀 ~ e.data:", e.data)
	const jsonData = JSON.parse(e?.data || {});
	// 处理返回结果
	renderResult(e?.data || {});

	if (jsonData.code === 0 && jsonData.data.status === 2) {
		setVoiceText(resultText.current);
		ws.close();
	}
	if (jsonData.code !== 0) {
		ws.close();
	}
};

ws.onerror = (e) => {
	// an error occurred
	console.log("🚀 ~ e.error", e.message)
};

ws.onclose = (e) => {
	// connection closed
	console.log("🚀 ~ e.code:", e.code, e.reason)
};

3、处理语音返回结果

const resultText = React.useRef<string>('');
const resultTextTemp = React.useRef<string>('');
// 识别结束 处理返回结果
const renderResult = (resultData: string) => {
	let jsonData = JSON.parse(resultData);
	if (jsonData.data && jsonData.data.result) {
		let data = jsonData.data.result;
		let str = "";
		let { ws } = data;
		for (let i = 0; i < ws.length; i++) {
			str = str + ws[i].cw[0].w;
		}
		// 开启wpgs会有此字段(前提：在控制台开通动态修正功能)
		// 取值为 "apd"时表示该片结果是追加到前面的最终结果；取值为"rpl" 时表示替换前面的部分结果，替换范围为rg字段
		if (data.pgs) {
			if (data.pgs === "apd") {
				// 将resultTextTemp同步给resultText
				resultText.current = resultTextTemp.current;
			}
			// 将结果存储在resultTextTemp中
			resultTextTemp.current = resultText.current + str;
		} else {
			resultText.current = resultText.current + str;
		}
	}
}