Start Here: LiveKit & Krisp
30 entriesResources for the two platforms behind this page: LiveKit for building agents and Krisp for voice quality.
Open-Source Voice Agent Frameworks
19 entriesFrameworks and building blocks for assembling a voice agent yourself.
Voice Agent Platforms & Realtime APIs
17 entriesHosted platforms and model-provider APIs for voice agents.
Speech Models: STT, TTS & Speech-to-Speech
34 entriesModels and comparisons for hearing, speaking and conversing.
Audio Quality & Voice Activity Detection
6 entriesCleaning audio and detecting speech before it reaches the model.
Telephony & Real-Time Infrastructure
8 entriesPhone networks, SIP and WebRTC plumbing that connects agents to callers.
Testing, Evaluation & Learning
7 entriesTools and guides for checking that a voice agent behaves.
Related Technologies on This Site
11 entriesAgent, chatbot, speech analytics and observability pages on this site.
Frequently Asked Questions
What is a voice agent?
A voice agent is an AI agent you talk to. It listens with speech recognition, reasons with a language model, speaks with text-to-speech, and can call tools or take actions, usually over a phone line or a web or mobile app. Some newer systems use speech-to-speech models that skip the separate text step.
What are the main parts of a voice agent?
Common components are speech-to-text, a language model, text-to-speech, voice activity detection and turn detection to know when the person has finished speaking, and a transport layer such as WebRTC or SIP telephony. LiveKit Agents and Pipecat are open-source frameworks that wire these together.
What do LiveKit and Krisp do?
LiveKit is an open-source framework and developer platform for building, testing, deploying and observing realtime voice, video and physical AI agents. Krisp provides noise cancellation and other voice AI technology, and its developer SDK is aimed at voice agents and apps. Krisp has described its voice-isolation and turn-prediction products for agents, and its performance figures are vendor-reported.
Why do turn detection and noise handling matter?
Agents that mistake background sound for speech interrupt people or lose the thread, and ones that wait too long feel slow. Turn detection decides when a speaker is done, and noise cancellation and voice activity detection improve what the model hears. LiveKit documents its turn handling, and Pipecat has an open Smart Turn model.
How do I test a voice agent?
Test with simulated conversations before launch and score real calls afterward. LiveKit offers an agent console, a CLI debugger and unit tests, and tools such as Hamming, Coval, Cekura and Roark simulate calls and evaluate production conversations.