plivo-stream-sdk-node

A Node.js SDK for handling Plivo real-time media streaming over WebSocket. Built on top of ws WebSocketServer.

Installation

npm install plivo-stream-sdk-node
# or
bun install plivo-stream-sdk-node

Quick Start

import express from 'express';
import PlivoWebSocketServer from 'plivo-stream-sdk-node';
import type { StartEvent, MediaEvent, DTMFEvent } from 'plivo-stream-sdk-node';
import * as Plivo from 'plivo';

const app = express();
const PORT = 8000;

// Plivo webhook endpoint - returns XML to initiate streaming
app.get('/stream', (req, res) => {
  const streamUrl = `wss://${req.get('host')}/stream`;
  const plivoResponse = new (Plivo as any).Response();
  plivoResponse.addSpeak('Hello world!');
  const params = {
    contentType: 'audio/x-mulaw;rate=8000',
    keepCallAlive: true,
    bidirectional: true,
  };
  plivoResponse.addStream(streamUrl, params);
  res.header('Content-Type', 'application/xml');
  res.header('Content-Length', plivoResponse.toString().length.toString());
  res.header('Connection', 'keep-alive');
  res.header('Keep-Alive', 'timeout=60');
  const xml = plivoResponse.toXML();
  res.type('application/xml');
  res.send(xml);
});

// Start HTTP server
const server = app.listen(PORT, () => {
  console.log(`Server listening on http://localhost:${PORT}`);
});

// Create PlivoWebSocketServer attached to your HTTP server
const plivoServer = new PlivoWebSocketServer({ server, path: '/stream' });

plivoServer
  .onConnection(async (ws, req) => {
    console.log('New WebSocket connection');
    // Initialize per-connection resources here (e.g., speech-to-text clients)
  })
  .onStart((event: StartEvent, ws) => {
    console.log('Stream started:', event.start.streamId);
    console.log('Call ID:', event.start.callId);
    console.log('Media format:', event.start.mediaFormat);
  })
  .onMedia((event: MediaEvent, ws) => {
    // Get raw audio buffer from the event
    const audioBuffer = event.getRawMedia();
    // Process audio (e.g., send to speech-to-text service)
  })
  .onDtmf((event: DTMFEvent, ws) => {
    console.log('DTMF digit:', event.dtmf.digit);

    // Example: clear audio queue on * press
    if (event.dtmf.digit === '*') {
      plivoServer.clearAudio(ws);
    }
  })
  .onPlayedStream((event) => {
    console.log('Stream played:', event.name);
  })
  .onClearedAudio((event) => {
    console.log('Audio cleared:', event.streamId);
  })
  .onError((error, ws) => {
    console.error('Stream error:', error.message);
  })
  .onClose((ws) => {
    console.log('Connection closed');
  })
  .start(); // Must call .start() to begin accepting connections

API Reference

PlivoWebSocketServer

Extends WebSocketServer from the ws package.

Constructor

new PlivoWebSocketServer(options: ServerOptions, callback?: () => void)

Standard ws ServerOptions. Common options:

• server: HTTP/HTTPS server to attach to
• path: URL path for WebSocket connections (e.g., '/stream')
• port: Port to listen on (if not attaching to existing server)

Lifecycle Methods

start(): this

Start accepting WebSocket connections. Must be called after registering all event handlers.

close(callback?: () => void): void

Close the WebSocket server.

Event Registration Methods (Chainable)

All return this for chaining. Multiple handlers can be registered per event.

Method	Callback Signature	Description
onConnection	(ws, request) => void | Promise<void>	New connection established. Async callbacks are awaited before processing messages.
onStart	(event: StartEvent, ws) => void	Stream initialization with call metadata
onMedia	(event: MediaEvent, ws) => void	Incoming audio chunk
onDtmf	(event: DTMFEvent, ws) => void	DTMF digit received
onPlayedStream	(event: PlayedStreamEvent, ws) => void	Audio playback confirmation
onClearedAudio	(event: ClearedAudioEvent, ws) => void	Audio queue cleared confirmation
onError	(error: Error, ws) => void	Error occurred
onClose	(ws) => void	Connection closed

Action Methods

playAudio(ws, contentType, sampleRate, payload)

Send audio to a specific connection.

// payload can be Buffer, Uint8Array, or ArrayBuffer
plivoServer.playAudio(ws, 'audio/x-mulaw', 8000, audioBuffer);

checkpoint(ws, name)

Send a checkpoint event to track audio playback progress.

plivoServer.checkpoint(ws, 'greeting-complete');

clearAudio(ws)

Clear all queued audio for a connection.

plivoServer.clearAudio(ws);

Getter Methods

Method	Return Type	Description
getStreamId(ws)	string | undefined	Stream ID for the connection
getAccountId(ws)	string | undefined	Plivo account ID
getCallId(ws)	string | undefined	Call ID
getHeaders(ws)	string | undefined	Extra headers from start event
isActive(ws)	boolean	Whether connection is open

Event Types

StartEvent

{
  event: 'start';
  sequenceNumber: number;
  start: {
    callId: string;      // UUID
    streamId: string;    // UUID
    accountId: string;
    tracks: string[];
    mediaFormat: {
      encoding: string;
      sampleRate: number;
    };
  };
  extra_headers: string;
}

MediaEvent

{
  event: 'media';
  sequenceNumber: number;
  streamId: string;
  media: {
    track: string;
    timestamp: string;
    chunk: number;
    payload: string;  // base64 encoded audio
  };
  extra_headers: string;
  getRawMedia(): Buffer;  // Helper to decode payload
}

DTMFEvent

{
  event: 'dtmf';
  sequenceNumber: number;
  streamId: string;
  dtmf: {
    track: string;
    digit: string;
    timestamp: string;
  }
  extra_headers: string;
}

PlayedStreamEvent

{
  event: 'playedStream';
  sequenceNumber: number;
  streamId: string;
  name: string;
}

ClearedAudioEvent

{
  event: 'clearedAudio';
  sequenceNumber: number;
  streamId: string;
}

Types Export

import type {
  StartEvent,
  MediaEvent,
  DTMFEvent,
  PlayedStreamEvent,
  ClearedAudioEvent,
  PlayAudioEvent,
  CheckpointEvent,
  ClearAudioEvent,
  IncomingEventEnum,
  OutgoingEventEnum,
} from 'plivo-stream-sdk-node';

---

Running the Example App

The examples/express-streaming directory contains a complete voice AI example using:

• Deepgram - Real-time speech-to-text
• OpenAI - Chat completion for responses
• ElevenLabs - Text-to-speech

Prerequisites

1. Node.js 18+ or Bun
2. A Plivo account with streaming enabled
3. API keys for Deepgram, OpenAI, and ElevenLabs
4. A way to expose your local server (e.g., ngrok)

Setup

1. Navigate to the example directory:

cd examples/express-streaming

2. Install dependencies:

npm install
# or
bun install

3. Create a .env file:

PORT=8000

# Deepgram (https://console.deepgram.com)
DEEPGRAM_API_KEY=your_deepgram_api_key
DEEPGRAM_MODEL=nova-2

# OpenAI (https://platform.openai.com)
OPENAI_API_KEY=your_openai_api_key
OPENAI_MODEL=gpt-4o-mini

# ElevenLabs (https://elevenlabs.io)
ELEVENLABS_API_KEY=your_elevenlabs_api_key
ELEVENLABS_VOICE_ID=your_voice_id
ELEVENLABS_MODEL_ID=eleven_turbo_v2

4. Start the server:

npx ts-node server.ts
# or with bun
bun run server.ts

5. Expose your server using ngrok:

ngrok http 8000

6. Configure Plivo:

   • Go to your Plivo console
   • Set your application's Answer URL to: https://your-ngrok-url.ngrok.io/stream

7. Make a call to your Plivo number and start talking!

How It Works

1. When a call comes in, Plivo hits the /stream endpoint
2. The XML response initiates a bidirectional WebSocket stream
3. Audio from the caller is sent to Deepgram for transcription
4. Transcriptions are sent to OpenAI for a response
5. OpenAI's response is converted to speech via ElevenLabs
6. The audio is streamed back to the caller in real-time

DTMF Controls

• Press * to clear the audio queue (interrupt the AI)

---

License

MIT