v0.1.0

First public release of the VoxRT wake-word iOS Swift Package — detects the phrase "Hey Assistant" on the VoxRT custom on-device inference runtime.

Quality

Held-out test split: 5,240 positive utterances + 6,416 hard-negative utterances. Speakers disjoint from train + val.

• ROC AUC: 0.9966
• Average precision (PR AUC): 0.9899

At the recommended deploy threshold of 0.9: precision 0.993, recall 0.982, F1 0.987, FPR 0.5 %.

Runtime performance

arm64 device build, post-warmup, RTF = wall-time-per-frame ÷ frame audio duration:

Device	SoC	RTF
iPhone 13 Pro Max	Apple A15 Bionic	0.015

≈ 65× faster than realtime — well within an always-on power budget.

Binary footprint

• VoxrtWakeWordNative.xcframework.zip (this asset): ~19 MB compressed (device + simulator slices)
• After SPM extraction + linker dead-code elimination on the device-only path: ~2–3 MB delta in your app binary
• Wake-phrase model voxrt_wake_word.vxrt: ~100 KB fp16 (downloaded separately)

Install

In Xcode: File → Add Package Dependencies → paste:

https://github.com/VoxRT/voxrt-wake-word-ios

…and pin to v0.1.0.

The wake-phrase model is NOT bundled — fetch from voxrt-wake-word-models v0.1.0.

License

VoxRT proprietary binary (xcframework) — redistribution as part of the unmodified package is permitted for commercial apps without per-installation
fees. See LICENSE-BINARY for full terms. Custom phrases via the commercial
VoxRT SDK — contact help@voxrt.com.