Screenshots 2
VoxRT Wake-Word
Summary
Most cloud voice SDKs bill per detection, send audio to a server, and add 200–400ms of round-trip latency — which makes them a non-starter for always-on embedded devices and privacy-regulated products. VoxRT is an on-device SDK that puts the wake-word engine, keyword spotter, and speech-to-intent model inside the binary, no network call required.
The SDK ships a Rust runtime under 1 MB with wake-word models around 100 KB, so it fits on mobile and IoT targets without gutting your memory budget. Audio stays on the device — the vendor states models are encrypted at rest and the system works offline by default, which means GDPR and HIPAA conversations get simpler, not harder. The published models are free for commercial use; custom models trained to your phrase, accent profile, or domain vocabulary are a paid engagement. iOS and Android are available in v1; Windows, WebAssembly, microcontrollers, automotive, and wearables are listed as v2, meaning shipping on those targets today is not an option. Teams that need a language other than English are also waiting — multilingual support is post-v1 on the roadmap.
Bottom line: Pick VoxRT for a privacy-first iOS or Android app that needs 'Hey YourBrand' to work in a tunnel — plan a different path if your device is a Cortex-M4 or your users speak Spanish.
Pricing Plans
Free- Free Tier
- Published models free for commercial production use with no per-user or per-detection fees
Published Models
Free for commercial use on supported platforms
- Wake word 'Hey Assistant'
- VAD
- Streaming ASR
- No fees
Custom Models
Paid engagements for custom phrases, vocabularies, and domains
- Tuned to spec
- Additional languages
- OEM support
View full pricing on voxrt.com →
Pricing may have changed since last verified. Check the official site for current plans.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Pros
Sign in to edit- Runtime under 1 MB with wake-word models around 100 KB, so the SDK fits on memory-constrained mobile and embedded targets where competing runtimes cannot be installed.
- No cloud round-trip and no per-detection fees, which means always-on listening stays within battery and cost budgets that would make a cloud-dependent architecture unshippable.
- Audio never leaves the device and models are encrypted at rest, so voice features pass privacy and compliance reviews that would block any SDK sending audio to a third-party server.
- Voice activity detection gates the heavier models, so the battery drain of continuous microphone monitoring is cut to the minimum — critical for wearables and IoT where always-on is the use case.
- Published models are free for commercial use with no account required, so a team can validate accuracy on real hardware before committing to a paid custom-model engagement.
Cons
Sign in to edit- Microcontroller targets — ARM Cortex-M4, M7, M33, M55, M85 — are listed as v2 and not available. Teams building firmware for these chips today cannot use VoxRT and will need a competitor like Picovoice Porcupine or Arm's ML Embedded Evaluation Kit, which already ship no_std-compatible binaries.
- English is the only supported language in v1. A product shipping to Spanish or French-speaking markets has no path forward with VoxRT until post-v1 multilingual support lands — no timeline is stated on the vendor page.
- Custom model training — tuning the wake phrase to your brand name, accent distribution, or noise profile — is a paid vendor engagement, not a self-service pipeline. Teams that expected to iterate on model accuracy independently will find themselves dependent on VoxRT's turnaround cycle for each training run.
- Windows and WebAssembly support is v2, meaning browser-based demos and Windows desktop apps cannot ship with VoxRT in v1. Teams prototyping on the web before committing to a mobile build lose the ability to test the actual SDK in that environment.
About
- Platforms
- iOS 16+, Android 8.0+, Linux, macOS, Windows, microcontrollers (ARM Cortex-M), Raspberry Pi, Jetson
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-09-09T00:03:00.170Z
Best For
Who it's for
- Embedded and IoT devices
- Mobile apps requiring offline voice
- Privacy-focused voice features
- Low-latency always-listening applications
- Custom phrase detection across accents and noise
What it does well
- Custom wake word activation like 'Hey YourBrand'
- Real-time on-device speech transcription
- Voice command keyword spotting
- Always-on voice activity detection for battery efficiency
- End-to-end speech-to-intent without transcripts
Integrations
Add notes, reviews, and benchmarks so the next visitor gets a clearer picture.
Compare VoxRT Wake-Word
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeFrequently Asked Questions
- Is VoxRT Wake-Word free?
- VoxRT Wake-Word has a permanent free tier alongside paid upgrades. You can keep using a baseline version indefinitely without paying.
- Is VoxRT Wake-Word open source?
- No — VoxRT Wake-Word is a closed-source tool. Source code is not publicly available.
- Can I self-host VoxRT Wake-Word?
- Yes. VoxRT Wake-Word supports self-hosting on your own infrastructure.
- When was VoxRT Wake-Word released?
- VoxRT Wake-Word was first released in 2026.
- What platforms does VoxRT Wake-Word support?
- VoxRT Wake-Word is available on: iOS 16+, Android 8.0+, Linux, macOS, Windows, microcontrollers (ARM Cortex-M), Raspberry Pi, Jetson.
Best VoxRT Wake-Word alternatives →
Curated lists that include this category
Cloud voice SDKs create friction for always-on devices
Most cloud voice SDKs bill per detection, send audio to a server, and add 200–400ms of round-trip latency — which makes them a non-starter for always-on embedded devices and privacy-regulated products. VoxRT Wake-Word ships an on-device SDK that keeps the wake-word engine, keyword spotter, and speech-to-intent model inside the binary with no network call required.
Runtime and privacy
The SDK includes a Rust runtime under 1 MB and wake-word models around 100 KB. The vendor states models are encrypted at rest and the system works offline by default. Audio never leaves the device. Published models are free for commercial use; custom models are a paid engagement.
Current reach and limits
iOS and Android ship in v1. Windows, WebAssembly, microcontrollers, automotive, and wearables are listed as v2. English is the only supported language in v1. Use cases include custom wake-word activation, on-device transcription, keyword spotting, voice activity detection, and end-to-end speech-to-intent.
Who it is for / who should skip it
Best for embedded and IoT devices, mobile apps needing offline voice, privacy-focused features, and low-latency always-listening. Skip it if you need microcontroller support today or non-English languages in the current release.