Documentation
Free API Key

Build a Voice Agent with Twilio & OpenAI & Deepgram

Learn how to build a voice agent application that includes Twilio, OpenAI, Deepgram speech-to-text & text-to-speech, while optimizing for low latency.

Suggest Edits

Integrating Twilio with Deepgram's Speech-to-Text (STT) and Text-to-Speech (TTS) functionalities allows you to build robust applications capable of real-time transcription and voice synthesis. This guide walks you through setting up a streaming voice agent using Twilio, Deepgram, and OpenAI, while optimizing for low latency.

Before You Begin

This guide assumes that you have a basic understanding of JavaScript, Node.js and are familiar with OpenAI, Twilio and Ngrok. It's also useful if you have general knowledge of how Large Language Models (LLMs) work.

📘

You can find the final code in The GitHub Repository.

📘

Before you can use Deepgram, you'll need to create a Deepgram account. Signup is free and includes $200 in free credit and access to all of Deepgram's features!

Create a Deepgram API Key

To access Deepgram’s API, you'll need to create a Deepgram API Key. Make note of your API Key; you will need it later.

Get Twilio Credentials

This demo uses Twilio Voice to start a phone call that will be recorded and transcribed. Before you can use Twilio's service, you'll need to sign up for a Twilio account.

To use the sample application, you'll need a Twilio Account SID and Twilio Auth Token. These can both be found within your Twilio Admin Dashboard.

Get OpenAI Credentials

This application uses OpenAI for it's LLM. Before you can use OpenAI's service, you'll need to sign up for an OpenAI account and obtain an API Key.

Getting Started

This is a basic JavaScript server that show cases end to end streaming voice agent with the following components:

  • Callable Phone Number
  • Streaming Twilio - Inbound Audio
  • Streaming Deepgram - Speech to Text
  • Streaming OpenAI - LLM
  • Streaming Deepgram - Text to Speech
  • Streaming Twilio - Outbound Audio

This implementation is a good starting reference to develop your own application logic and isn't design for scale or production deployments.

Clone the Repository

Either clone or download the GitHub repository to your local machine in a new directory:

# Clone this repo
git clone https://github.com/DamienDeepgram/deepgram-twilio-streaming-voice-agent.git

# Move to the created directory
cd deepgram-twilio-streaming-voice-agent

App sever setup

📘

Refer to the server.js file in the GitHub Repository to see the server side implementation.

Set environment variables

You will need to set environment variables for your shell session. These environment variables are used to store the API keys required for authentication when accessing the OpenAI and Deepgram APIs.

Open a Terminal and run:

export OPENAI_API_KEY=xxx
export DEEPGRAM_API_KEY=xxx

To verify the environment variables are set, run the following commands in your Terminal:

echo $OPENAI_API_KEY
echo $DEEPGRAM_API_KEY

Installation

Requires Node >= v12.1.0

Run npm install

Running the server

Start with npm run start

Demo Setup

Configure the environment using the Ngrok UI & CLI

  1. Install ngrok:
  1. Sign up for a ngrok account:
  • If you haven't already, sign up for an ngrok account
  • Copy your ngrok authtoken from your ngrok dashboard
  • Run the following command in your terminal to install the auth token and connect the ngrok agent to your account.
ngrok config add-authtoken <TOKEN>

Twilio Phone Number Setup

  • You will need to provide a valid phone number from Twilio.
  • You can either use the Twilio CLI (see instructions below) OR the Twilio Admin Dashboard to setup a phone number. (see these instructions from Twilio)

Configure with the Twilio CLI

Refer to this Repository for more information on this section.

  1. Install the Twilio CLI and login to Twilio to run these commands.

  2. Find an available phone number

twilio api:core:available-phone-numbers:local:list --country-code="US" --voice-enabled --properties="phoneNumber"`
  1. Purchase the phone number (where +123456789 is a number you found)
twilio api:core:incoming-phone-numbers:create --phone-number="+123456789"`

Set the webhook url to your ngrok url

  1. Start ngrok

On a separate terminal (not the one where you have run npm run start):

ngrok http 8080
  1. You will see a url under the Forwardingrow that --> to your localhost. Copy this as the <ngrok url>
  2. Edit the templates/streams file to replace <ngrok url> with your ngrok host. Example: wss://yourdomain.ngrok-free.app/streams. Remember to use wss://and include /streams in the URL.
  3. Go to your Twilio dashboard and select the active phone number that you manage. Under the Configure tab, include the \<ngrok urlin URL of the "A call comes in" row of the page. If your URL iswss://yourdomain.ngrok-free.app, put in <https://yourdomain.ngrok-free.app/twiml>

🚧

If you restart your ngrok server your ngrok url will change.

Call the number and chat to your bot.

  1. You can call the Twilio phone number directly from your own phone. Alternatively, you can make the call using the following, where +123456789 is the Twilio number you bought and +19876543210 is your phone number and abcdef.ngrok.io is your ngrok host.
twilio api:core:calls:create --from="+123456789" --to="+19876543210" --url="https://abcdef.ngrok.io/twiml"

Updated 3 months ago