barge-in

Addressee detection for voice agents: device-directed speech detection that runs before STT, so background speech, side conversations, and the agent's own TTS echo never trigger it. No wake word, model-agnostic, drop-in for LiveKit, Pipecat, ElevenLabs, Twilio, and OpenAI. The layer your VAD and turn detection are missing.

Self-hosted AI voice agent for Asterisk. Streaming STT to LLM to TTS over AudioSocket, with barge-in and tool calling.

Full-duplex, vision-enabled voice assistant for Windows — 100% local, on-device. Barge-in interrupts, desktop screenshots give the LLM visual context, and STT (whisper.cpp) + LLM (llama.cpp) run on Vulkan with no PyTorch/CUDA. Streaming Piper TTS in pt-BR. GGML edge AI.

Telnyx media-stream voice-agent reference implementation and a reproducible benchmark harness for measuring voice-agent interruption (barge-in) latency

Data Forge Hackathon Entry - Continuum, A voice-native conversation continuity assistant that delivers whispered recaps into your earpiece before you ever say 'hello'. - Team Byte Club

Local-first VAD, barge-in, and turn-taking primitives for interruptible voice agents.

Fast, interruptible browser voice agent: VAD → STT → LLM → TTS with real barge-in, echo filtering, streaming TTS and tap-to-seek. Zero runtime dependencies.

Measure what a caller actually hears from a telephony voice agent: frame pacing, continuity and barge-in latency over AudioSocket.

eval and regression harness for elevenlabs voice agents: audio scoring, llm judges, baseline diffs, closed-loop prompt refinement, ship verdicts

Realtime voice AI assistant — streaming STT → LLM → TTS over WebSocket, with barge-in and live translation. Runs fully local (Whisper + Ollama + Piper) or with cloud providers.

TypeScript starter kit for production voice agents: WebRTC, telephony, realtime model APIs, STT/TTS pipelines, barge-in, handoff, evals, and post-call workflows.

Full-duplex conversational turn-taking in C++17 — no ASR, no LLM, no neural VAD. Decides when you have finished speaking from prosody alone, and hears you interrupt it through its own speaker echo.

FastMCP 3.2 speech gateway: FunASR local STT, Gemini Live/TTS, Hume, ElevenLabs cloning, wake word, RAG. MCP server + React webapp.

Always-on spoken assistant. Wake with 헤이 자비스. Attach any agent brain.

A voice-first assistant that runs entirely on your own machine. Wake word, barge-in with real echo cancellation, memory that survives restarts, twenty tools, and a neural voice — no API keys, nothing uploaded. ~500ms from you stopping to it speaking, on an 8GB M1.

Know when your voice agent is still speaking. Server-side playback timeline for streamed audio, with barge-in positions and stall detection.

Autonomous real-time voice agent: streaming STT to LLM (function calling) to TTS with sub-500ms latency and barge-in. Offline-first, zero API keys.

Minimal real-time voice agent on the OpenAI Realtime API — streaming PCM, server-side VAD, function-calling into local RAG, browser + CLI demos with barge-in

Got tired of paying per minute to give an app a voice. Had a homeserver doing nothing, so I tried it with local models only — WebRTC, Whisper, Kokoro, Silero, all self-hosted. Works well for me. Not a finished product, take it and make it yours.