audio.cpp ships real-time local avatar voice chat: ~850ms latency on RTX 5090
Acceptable-Cycle4645 · reddit · 2026-11-16
audio.cpp 0.9.5 adds a real-time local avatar voice chat app in its WebUI, combining SoulX-FlashHead (talking-head generation), OWSM v4 (ASR), and LFM3-Audio (conversation + TTS, called twice for a second voice). 850ms question-to-speech latency on RTX 5090; cuDNN is mandatory (without it, 2x slower inference and doubled VRAM); SoulX needs RTX 4090+ for the GPU-heavy video stage.
More from Infra
- Local real-time talking-head voice chat lands in audio.cpp 0.9.4 with ~850ms latency on RTX 5090 — Acceptable-Cycle4645 · 2026-11-16
- Exo Buys Two More DGX Sparks to Push Local Inference Amid Open-Weights Ban Fears — kuchaev · 2026-11-16
- Bessent questions on TV whether Nvidia pressed "seemingly illicit sales" to boost revenue — ShakeelHashim · 2026-11-16
- AI 'Picks and Shovels' Trade Unravels as Data Center-Linked Stocks Lead Sell-Off: FT — bigdata · 2026-11-16
- Bittensor Pitches Itself as the Payment Layer for Open AI: 128 Subnets, No CEO — bittingthembits · 2026-11-16
- cudamemcpy-tuner speeds up pageable cudaMemcpy 5x on Nvidia B300 — makneeee · 2026-11-16