Outbound Twilio Call using ElevenLabs Voice/Agent

Help me Build my Workflow
1

Looking through the Twilo and ElevenLabs documentation there is a way to make outbound calls with Twilio while using a voice from ElevenLabs.

Two of the requirements I don’t know if they can be or should be in n8n selfhosted I have or if I need to host it on another container or not.

  • Node.js 16+
  • ngrok for local development.

I have a Google Sheets document with the records to call and Id like the workflow to call the numbers one by one and have a ending webhook call to route the returned information from the call to route to a seperate sheet.

Has anyone gotten Twilio outbound calls with ElevanLabs voice yet?

2

Hi, I have a similar use case and I’m very interested to know if this can be done.

3

The hangup for me is the part about needing Node.js 16+ and the files needed from the conversational-ai-twilo clone from github… and for ngrok… ive yet to get trafix and docker with the ssl stuff working just yet but I can see how ngrok is just built into the docker stack easily enough with my own domain or their free one.

4

Well, I found the Elevanlabs nodes, it does include the below… just the whole conversational agent using a voice thing isn’t there.

  • Text to speech: Turn any text into a high quality audio, lifelike, context-aware speech audio file that you can manipulate in your workflow.
  • Speech to speech: Transform your voice into an other character, preserving emotions and deliver with consistency.
  • Generate voice: Generate a new voice based on settings like age, gender, accent… add it to your library and start using it your workflow.
  • Clone your voice: Clone your own voice from an audio file and use it in your workflow, for text to speech or speech to speech.

I did notice how

5

really shocked it appears no one has implemented this…

6

Agreed. Surprising!

1 Like
7

I tried.. will post with how far I got to see if it helps get us all closer.

9

As far as I understand, Vapi Ai handles all this in the background, inbound and outbound. It connects quite easily with Twilio as well. But you pay for the convenience.

Isn’t the point of using Twilio/ElevenLabs directly for outbound to cut costs? So then, incorporating Vapi defeats that purpose.

10

I will look into this Vapi API, thanks!

11

It adds an extra latency, requests for each service will be the drawback here…

12

You are totally right here.

Ended up only with Twilio and n8n . At the end, it s all about API calls… no yet there a “super” platform that offers all the services under the same roof.

1 Like
13

I am working on a similar requirement where I am automating a whole interview process. I have managed to get the question - answer loop running. I am using Twilio for making calls and voice(just asking question), Deepgram for STT, OpenAI node for conducting the interview(analyzing answer,deciding relevance of answer and next action).

The interview is very much robotic and monotonous.I am working on making it more conversational and human like. Has anyone attempted any such use case? Please let me know if there is any better approach for this.

1 Like
14

Using Twilio TTS makes the voice feel robotic …

I have a simple example that I use this approach :

-predefined questions that are served directly with mp3 files made with 11labs for my own voice .

- gathering all the info from user and STT and if there are negative/positive/neutral responses you can prepare mp3 in advance.

All in Twilio, no other external requests(latency unde 1 second.

If you try to use LLM then for each API call, you get an extra latency, and that’s why feel as well robotic with delay +2 seconds responses…

1 Like
15

If your questions are always the same, this is a nice approach. I do want variable questions however, and then it won’t work.

16

Then I strongly suggest going into Python , OpenAi-RealTime, MediaStream Twilio. (Use websockets and you will get a Speech-to-Speech conversation) and keep n8n for orchestration or do it entirely in n8n, but since latency here is the key play role… :slight_smile:

1 Like
17

Were you able to automate the interview? im doing the exact same thing and am running into an issue with the call back with the voice agent. could you give me a few pointers on what you did?