VoxRT's profile on Product Hunt

About

VoxRT is an on-device voice AI runtime built in Rust from scratch. No ONNX, no TFLite. A NEON-tuned engine for ARMv8 and WASM SIMD128 that runs speech models locally on mobile, single-board computers, and in the browser. The stack: - Wake-word "Hey Assistant": ~100 KB, 5.3% RTF on a $15 Raspberry Pi Zero 2 W - Silero VAD (MIT) on our runtime: 3.05% RTF on a 2020 mid-range Android - Streaming ASR (NeMo FastConformer): real-time on mobile, WER 3.267% LibriSpeech - 14-command KWS: 16% RTF on Snapdragon 662, 0.97% in Chrome WASM SDKs are Apache-2.0 wrappers around a proprietary runtime. Custom phrases and vocabularies are a paid tier. Try in the browser: - https://voxrt.com/wake-word-demo - https://voxrt.com/asr-demo - https://voxrt.com/asr-demovoxrt.com/kws-demo

Links

Github

voxrt.com

Badges

Tastemaker

Gone streaking 10

Gone streaking

Gone streaking 5

Forums

p/voxrt-voice-ai-runtime

•

1mo ago

Four voice SDKs, one shared runtime, zero cloud

Quick update from the VoxRT team. On Sept 19 we are launching VoxRT Voice AI Runtime here on Product Hunt. Each model ships as its own SDK. All of them are built on the same VoxRT runtime. That keeps latency profiles, memory behavior, and platform bindings consistent across the stack.

Wake-Word shipped earlier this month as a standalone product. This launch introduces the rest of the SDKs alongside it.

•

24d ago

VoxRT Voice AI Runtime - The full voice stack, running on-device

No audio ever leaves the user's phone, browser, or Pi. Five SDKs, one shared runtime, zero cloud calls. Includes: wake-word "Hey Assistant" (100 KB, 5.3% RTF on Pi Zero), Silero VAD (3.05% RTF on Android), 14-command KWS (16% RTF on Snapdragon 662, 0.97% Chrome WASM), streaming ASR (real-time on mobile), and voice-cloning TTS (in development, coming soon). Runs on iOS, Android, browser (WASM), and Raspberry Pi. Coverage varies per component. No API keys, no per-minute pricing.

•

1mo ago

Where does voice actually fit in how people use AI?

Something we keep noticing: even when speaking a prompt would be faster, we still type it.

Voice feels great for quick asks like "set a timer" or "what's the weather" and awkward for anything that needs revising mid-sentence, or reading back a complex answer.