Transcription Server 1

speechToTextUtils.ts

import speech, { SpeechClient } from '@google-cloud/speech';
import { google } from '@google-cloud/speech/build/protos/protos';
import * as pumpify from 'pumpify';
import chalk from 'chalk';
import { Socket } from 'socket.io';
let speechClient: SpeechClient | null = null;

class SpeechToTextUtils {
	recognizeStream!: pumpify | null;
	resultEndTime = 0;
	isFinalEndTime = 0;
	finalRequestEndTime = 0;
	bridgingOffset = 0;
	streamingLimit = 290000;
	restartCounter = 0;
	lastTranscriptWasFinal = false;
	audioInput: DataView[] = [];
	lastAudioInput: DataView[] = [];
	newStream = true;
	socket!: Socket;
	request!: google.cloud.speech.v1.IStreamingRecognitionConfig | undefined;
	restartTimeout: NodeJS.Timeout | undefined;

	set _socket(value: Socket) {
		this.socket = value;
	}

	set _request(value: google.cloud.speech.v1.IStreamingRecognitionConfig) {
		this.request = value;
	}

	startRecognitionStream() {
		this.audioInput = [];
		if (!speechClient) {
			speechClient = new speech.SpeechClient(); // Creates a client
		}
		this.recognizeStream = speechClient
			.streamingRecognize(this.request)
			.on('error', (err) => {
				console.error('Error when processing audio: ' + err);
				this.socket.emit('googleCloudStreamError', err);
				this.stopRecognitionStream();
			})
			.on('data', this.speechCallback.bind(this));

		this.restartTimeout = setTimeout(
			this.restartStream.bind(this),
			this.streamingLimit
		);
	}

	speechCallback(stream: google.cloud.speech.v1.StreamingRecognizeResponse) {
		// Null checks
		if (
			stream.results &&
			stream.results[0] &&
			stream.results[0].resultEndTime &&
			stream.results[0].resultEndTime.nanos &&
			stream.results[0].resultEndTime.seconds &&
			stream.results[0].alternatives &&
			stream.results[0].isFinal
		) {
			// Convert API result end time from seconds + nanoseconds to milliseconds
			// The below seconds are useful to see the timestamps in the console
			let seconds: number;
			if (typeof stream.results[0].resultEndTime.seconds === 'string')
				seconds = parseInt(stream.results[0].resultEndTime.seconds);
			else if (Long.isLong(stream.results[0].resultEndTime.seconds))
				seconds = stream.results[0].resultEndTime.seconds.toNumber();
			else seconds = stream.results[0].resultEndTime.seconds;
			this.resultEndTime =
				seconds * 1000 +
				Math.round(stream.results[0].resultEndTime.nanos / 1000000);

			// Calculate correct time based on offset from audio sent twice
			const correctedTime =
				this.resultEndTime -
				this.bridgingOffset +
				this.streamingLimit * this.restartCounter;

			process.stdout.clearLine(0);
			process.stdout.cursorTo(0);
			let stdoutText = '';
			if (stream.results[0] && stream.results[0].alternatives[0]) {
				stdoutText =
					correctedTime + ': ' + stream.results[0].alternatives[0].transcript;
			}

			if (stream.results[0].isFinal) {
				process.stdout.write(chalk.green(`${stdoutText}\n`));
				this.socket.emit(
					'speechData',
					stream.results[0].alternatives[0].transcript
				);

				this.isFinalEndTime = this.resultEndTime;
				this.lastTranscriptWasFinal = true;
			} else {
				// Make sure transcript does not exceed console character length
				if (stdoutText.length > process.stdout.columns) {
					stdoutText =
						stdoutText.substring(0, process.stdout.columns - 4) + '...';
				}
				process.stdout.write(chalk.red(`${stdoutText}`));

				this.lastTranscriptWasFinal = false;
			}
		}
	}
}

export default new SpeechToTextUtils();