Omni-VRAM

RELEASE v2.7.1
High-Performance Full-Duplex Voice Runtime & Paged KV-Cache Engine for Edge Acceleration
PyPI Repository GitHub Source
Tests: 979 Passed / 0 Failed
979 / 0
Regression Suite (100% Pass)
< 150 ms
Time to First Audio (TTFA)
40%+
Paged Memory Compression
Production Integration
BASH (PYPI INSTALLATION) 
pip install --upgrade omni-vram
REALTIME WEBSOCKET GATEWAY RUNNER 
python -m vram_core.api_server --port 8000
Pipeline Topology
Microphone In (PCM16)
AEC NCC Veto
Whisper Stream
Paged KV LLM
Streaming TTS
Speaker Out
Interactive Arbiter: Acoustic Echo Veto (NCC) Threshold: 0.55

Real-time cross-correlation calculation between speaker reference buffer and microphone input to eliminate acoustic loop self-interruptions in hands-free environments.

Speaker Reference Energy (Ref RMS) 0.75
Normalized Cross-Correlation Peak (NCC R_max) 0.68
Correlation peak 0.68 ≥ 0.55 threshold: Identified as speaker feedback loop. TTS output stream protected.
VETOED (MUTE)