Spaces:
Running
Running
Personal project
#1
by
dvalle08 - opened
What makes this cool:
Most people think you need OpenAI's Realtime API or ElevenLabs to build a decent voice agent. This project proves you can do it with 100% open-source models running on consumer hardware.
The Stack:
Moonshine STT (61M params) - proves small models can do real-time transcription
Qwen2.5-3B-Instruct - shows you don't need 70B parameter monsters for conversation
PocketTTS - demonstrates natural voice synthesis (this runs in CPU)
LiveKit - professional WebRTC infrastructure
8GB VRAM is enough for real-time voice AI
space:
https://huggingface.co/spaces/dvalle08/open-voice-agent
dvalle08 changed discussion title from
Apply for a GPU community grant: Personal project
to Personal project