Three open-source models combine into a fully local, real-time voice-to-avatar assistant
shinjiw_at_cmu · x · 2026-11-06
A developer combined three independently developed open-source models into a real-time voice-to-avatar assistant running entirely locally, built on audio.cpp:
- LFM3-Audio-2B (Liquid AI), optimized for on-device inference, generates conversational responses and streams speech;
- SoulX-FlashHead animates a portrait from incoming audio chunks faster than real time;
- OWSM v4 (WavLab) handles multilingual ASR.
The three models work together as one live agent rather than an offline clip, with a demo of the complete pipeline released.
More from coding & agent
- Jev 公开 72 万条决策数据,可自训轻量级路由/决策模型 — MaziyarPanahi · 2026-11-06
- Claude + Blender MCP produces stunning abandoned food court 3D scene — sidahuj · 2026-11-06
- CRUX-in-a-box: open-source infra for weeks-long, monitored agent evaluations — sayashk · 2026-11-06
- Can anyone name one production app that is 90-100% AI-agent generated? — Zamdi · 2026-11-06
- Jensen Huang: engineers using agents are busier, not less, because they must verify output — shashib · 2026-11-06
- E2B hits $120M ARR, credits 'Agent-Led Growth' for B2B devtool sales — badphilosopher · 2026-11-06