Local voice agent on a 2015 laptop hits 20-30s latency — advice needed
mesolimbic-bliss · reddit · 2026-10-27
Inspired by the Reachy-Mini robot, a user tried building a snarky local voice companion on an old MSI GS70 laptop (i7-5700HQ, 16GB DDR3, GTX 965M with 2GB VRAM) using Open WebUI + llama.cpp + Gemma 4 E2B with Whisper STT + Kokoro TTS. All attempts show 20-30 second latency. They're asking for optimization advice — no avatar needed, just a voice to keep them company during late-night work. Question: is the laptop simply too old and slow for a tiny voice model?
More from Infra
- Dragonfruit runs its ventures on Apple Silicon with fixed-fee open-weight inference — akumar · 2026-10-28
- Token pricing has no ceiling: agents burn compute without breaks, reshaping enterprise AI billing — akumar · 2026-10-28
- Vjeux's AI agents hit 2.6x speedups on Open Kernel Challenge kernels, then stalled — Vjeux · 2026-10-28
- "My job is to make tokens as cheap as humanly possible" — every chip, every watt, every acre — GregKamradt · 2026-10-28
- General Compute raises $100M Series A to bet on fast inference demand — _akhaliq · 2026-10-28
- Microsoft's custom Maia chips to make up only ~10% of its AI chip spend by early 2027, says Evercore — Beth_Kindig · 2026-10-28