136 lines
3.8 KiB
TypeScript

// import WebSocket from 'ws';
import { initWs } from '../../../ws-adapter/index.ts';
import { logger } from '@/logger/index.ts';
import { WSServer, WSSOptions } from '../../provider/ws-server.ts';
export type VideoWSOptions = {
url?: string;
ws?: WebSocket;
itn?: boolean;
mode?: VideoWsMode;
isFile?: boolean;
onConnect?: () => void;
wav_format?: string;
} & {
wsOptions?: WSSOptions['wsOptions'];
};
export const videoWsMode = ['2pass', 'online', 'offline'] as const;
type VideoWsMode = (typeof videoWsMode)[number];
type OpenRequest = {
// 语音分片大小(单位: 毫秒):
chunk_size: number[];
// 音频文件名:
wav_name: string;
// 是否正在说话:
is_speaking: boolean;
// 分片间隔(单位: 毫秒):
chunk_interval: number;
// 逆文本标准化(ITN):
itn: boolean;
// 模式:
// '2pass' - 双通道模式, 'online' - 在线模式, 'offline' - 离线模式
mode: VideoWsMode;
// 音频格式:
wav_format?: string;
// 音频采样率(单位: Hz):
audio_fs?: number;
// 热词列表:
hotwords?: string;
};
export type VideoWsResult = {
isFinal: boolean;
mode: VideoWsMode;
stamp_sents: { end: number; punc: string; start: number; text_seg: string; tsList: [][] }[];
text: string;
timestamp: string;
wav_name: string;
};
export class VideoWS extends WSServer {
itn?: boolean;
mode?: VideoWsMode;
wav_format?: string;
constructor(options?: VideoWSOptions) {
super({ url: options?.url, ws: options?.ws, onConnect: options?.onConnect, wsOptions: options?.wsOptions });
this.itn = options?.itn || false;
this.itn = options?.itn || false;
this.mode = options?.mode || 'online';
this.wav_format = options?.wav_format;
}
async start(opts?: Partial<OpenRequest>) {
const chunk_size = new Array(5, 10, 5);
const request: OpenRequest = {
chunk_size: chunk_size,
wav_name: 'h5', //
is_speaking: true,
chunk_interval: 10,
itn: this.itn,
mode: this.mode || 'online',
...opts,
};
const file_sample_rate = 16000;
request.wav_format = request.wav_format || this.wav_format || 'wav';
if ('wav' == request.wav_format) {
request.wav_format = 'PCM';
request.audio_fs = file_sample_rate;
}
console.log('request', request);
this.ws.send(JSON.stringify(request));
}
async stop() {
var chunk_size = new Array(5, 10, 5);
var request = {
chunk_size: chunk_size,
wav_name: 'h5',
is_speaking: false,
chunk_interval: 10,
mode: this.mode,
};
this.ws.send(JSON.stringify(request));
}
async send(data: any) {
if (this.ws && this.ws.readyState === WebSocket.OPEN) {
this.ws.send(data);
}
}
async sendBuffer(data: Buffer, opts?: { isFile?: boolean; wav_format?: string }) {
const { wav_format = 'wav' } = opts || {};
if (this.ws && this.ws.readyState === WebSocket.OPEN) {
let sampleBuf = new Uint8Array(data);
const ws = this;
var chunk_size = 960; // for asr chunk_size [5, 10, 5]
let totalsend = 0;
let len = 0;
ws.start({ wav_format });
while (sampleBuf.length >= chunk_size) {
const sendBuf = sampleBuf.slice(0, chunk_size);
totalsend = totalsend + sampleBuf.length;
sampleBuf = sampleBuf.slice(chunk_size, sampleBuf.length);
await new Promise((resolve) => setTimeout(resolve, 10));
ws.send(sendBuf);
len++;
}
ws.stop();
}
}
async onMessage(event: MessageEvent) {
super.onMessage(event);
const data = event.data;
try {
const result = JSON.parse(data.toString());
console.log('result', result);
} catch (error) {
console.log('error', error);
}
}
async onError(event: Event) {
console.log('onError', event);
}
async onClose(event: CloseEvent) {
console.log('onClose', event);
}
}