Skip to main content
Build a voice agent using Deepgram for speech recognition, Google Gemini for conversation, and ElevenLabs for natural-sounding text-to-speech synthesis. Best for: Balance of cost efficiency and voice quality with multilingual support.

Prerequisites


Installation


Environment Variables


Pipeline Configuration


Service Details

Deepgram STT

Real-time speech recognition with interim results and language detection.

Google Gemini LLM

Streaming responses with function calling and multimodal input support. Features:
  • Streaming responses
  • Function calling
  • Multimodal inputs (text, images)
  • OpenAI-compatible context format

ElevenLabs TTS

Natural voice synthesis with word-level timing and voice cloning support. Service options:
  • ElevenLabsTTSService - WebSocket-based, recommended for real-time
  • ElevenLabsHttpTTSService - HTTP-based, simpler setup

Quick Start

Inbound Calls

Configure your Plivo number’s Answer URL to your ngrok URL.

Outbound Calls


Pipecat Overview

Architecture and setup

Deepgram Docs

STT configuration

Gemini Docs

LLM configuration

ElevenLabs Docs

TTS configuration