Voice Agents: Real-Time Speech AI Pipelines

Explore 7 voice agent projects for real-time speech AI, including coaching, healthcare contact centers, and live conversational assistants.

The Voice Agents category includes 7 projects that demonstrate real-time speech processing, conversational voice AI, and streaming pipelines. These projects show how to build agents that listen, speak, and reason in real time, powered by frameworks such as LiveKit, Pipecat, Gradium, and Sarvam.

Voice Agent Projects

Project	Technologies	Description
Gradium + Nebius Voice Coach	Gradium, Nebius, LangChain	Conversational pitch coach using Gradium STT/TTS, LangChain orchestration, and Nebius reasoning
Healthcare Voice Contact Center	Pipecat, healthcare APIs	Contact center with appointment booking, FAQ handling, and supervisor escalation
LiveKit + Gemini Realtime	LiveKit, Google Gemini	LiveKit Agents with Gemini Live multimodal for low-latency voice conversations
LiveKit Voice Agent with Web Search	LiveKit, Gemini, Olostep	Live realtime voice agent with web search for fresh, source-cited answers
LiveKit RSVP Confirmation	LiveKit, JSON DB	Outbound voice agent that calls attendees, confirms RSVPs, and updates an event database
Pipecat + Sarvam	Pipecat, Sarvam, OpenAI	Voice pipeline with Sarvam STT/TTS and OpenAI chat over WebRTC or Daily transport
Speed-to-Lead Voice Agent	LiveKit, mock CRM	Voice agent that calls inbound leads instantly, routes them to specialists, and logs to CRM

Getting Started with Voice Agents

1. Navigate into a voice project

   cd voice_agents/<project_name>
   
2. Configure your environment

   cp .env.example .env
   

   Voice projects typically require STT/TTS keys (e.g., Deepgram, ElevenLabs, Sarvam) and LLM keys (e.g., OpenAI, Nebius, Gemini).

3. Install and run

   pip install -r requirements.txt
   python main.py
   

   Some projects may also expose a web UI for testing.

Voice projects often require WebSocket or WebRTC transport. For projects using LiveKit or Daily, follow the transport setup instructions in the individual READMEs.