voxrt

Streaming on-device speech recognition for Android — NEON-accelerated FastConformer (32M params), cache-aware streaming with 80 ms look-ahead, no cloud. Powered by the VoxRT runtime.

On device streaming voice activity detection (Silero VAD v5) for iOS. Custom Rust inference runtime, NEON-accelerated arm64, RTF ~1.85% on iPhone.

On device streaming voice activity detection (Silero VAD v5) for Android. ~424 KB native binary, NEON-accelerated arm64-v8a, RTF ~3% on Snapdragon 662.

On-device wake-phrase detection for Linux aarch64 — one glibc-2.17 .so behind Python (PyPI wheel), Node.js (npm), Go (go get), C/C++ (tarball + CMake), Rust. RTF 0.053 on Raspberry Pi Zero 2 W. Same runtime as voxrt-wake-word-{android,ios}.

Streaming on-device speech recognition for iOS — NEON-accelerated, encrypted FastConformer (32M params), RTF 0.08–0.10 on iPhone 13 Pro Max. Built on the VoxRT custom Rust inference runtime. SwiftPM distribution.

On-device 14-way keyword spotting in the browser — WebAssembly + SIMD128, ~1.5 MB total (runtime + model), no server. Same runtime as voxrt-kws-{linux,android,ios}.

Always-on wake-phrase detection for iOS on the VoxRT custom on-device inference runtime — Swift Package, 16 kHz mono PCM in, threshold-crossing events out. Custom phrases at voxrt.com.

Pre-compiled ASR model weights for the VoxRT on-device runtime. Encrypted .vxrt v2 format. streaming-medium-pc: FastConformer 32M, CTC + RNN-T, CC-BY-4.0 (NVIDIA NeMo).

On-device wake-phrase detection in the browser — WebAssembly + SIMD128, ~275 KB total (runtime + model), no server. Same runtime as voxrt-wake-word-{android,ios,linux}. Free demo tier of the VoxRT wake-word family.

Always-on wake-phrase detection for Android on the VoxRT custom on-device inference runtime — Kotlin library, 16 kHz mono PCM in, threshold-crossing events out. Custom phrases at voxrt.com.

On-device streaming ASR (speech-to-text) for Linux x86_64 + aarch64 — one hardened .so behind Python (PyPI), Node.js (npm), Go (go get), C/C++ (tarball + CMake). AVX2 on servers, NEON on edge. RNN-T + CTC, RTF 0.06–0.08 on x86_64. Same runtime as voxrt-asr-{android,ios}.

On-device 14-way keyword spotter for Linux x86_64 + aarch64 — one glibc-2.17 .so behind Python (PyPI), Node.js (npm), Go (go get), C / C++ (tarball + CMake). Streaming Conformer, RTF 1 % on WSL AVX2, 32 % on Pi Zero 2 W NEON. Same runtime as voxrt-kws-{android,ios} (upcoming).

Pre-compiled Silero v5 VAD weights in .vxrt format for the VoxRT inference runtime. AES-256-GCM encrypted, ~1.2 MB, MIT.

Wake-phrase model weights (.vxrt, ~100 KB, AES-GCM encrypted) for the VoxRT custom on-device inference runtime. Paired with voxrt-wake-word-{android,ios,linux}. Custom phrases at voxrt.com.

14-way keyword-spotting model weights (.vxrt v2, AES-256-GCM encrypted) for the VoxRT on-device runtime. Streaming Conformer-Medium (636 K params, 1.28 MB). Paired with voxrt-kws-{linux,browser}.

On-device 14-way keyword spotter for iOS — Swift Package on the VoxRT custom Rust runtime. 16 kHz mono PCM in, per-class detections out. Streaming Conformer with NEON kernels, same runtime as voxrt-kws-{linux,android,browser}. Upcoming.

On-device 14-way keyword spotter for Android — Kotlin library on the VoxRT custom Rust runtime. 16 kHz mono PCM in, per-class detections out. Streaming Conformer, RTF 16 % on Snapdragon 662 A73. Same runtime as voxrt-kws-{linux,browser}.

Stream on-device speech recognition on Android using the custom VoxRT inference runtime with NeMo FastConformer support.

Distribute pre-compiled VoxRT wake-word model weights for on-device inference using the compact .vxrt format.

Run Silero v5 voice activity detection on iOS using a custom, high-performance Rust inference runtime for low-latency streaming.