Mickai Subsystem
Mickai Jarvis™
Mickai Jarvis is the voice subsystem of the Mickai SIOS. Speaker verification against a hardware-bound voiceprint, fresh-utterance authority on high-impact actions, on-device speech recognition, on-device speech synthesis. Audio never leaves the machine. Mickai is downloadable at mickai.co.uk/download and runs on Windows, Linux, or macOS.
View capabilitiesThe Mickai SIOS
Mickai is a Sovereign Intelligence Operating System (SIOS). It runs entirely on your own hardware, on Windows, Linux, or macOS. No cloud, no telemetry. This page describes one subsystem of the Mickai SIOS. Download Mickai at mickai.co.uk/download.
A subsystem of the Mickai SIOS. Voice gating, voice-biometric authentication, and speech synthesis. Actor identity cannot be supplied by a public reply.
Actor identity cannot be supplied by a public reply.
What Jarvis does
Seven primitives that turn voice into a load-bearing authority signal. Every gate decision is signed into the Open Audit Record substrate. Audio is captured, verified, and rendered entirely on the host.
01 / Biometric
Voiceprint speaker verification
Jarvis verifies the speaker against a voiceprint sealed inside the secure enclave at enrolment. The voiceprint is never extractable. Verification compares the live utterance against the sealed template entirely on the host; no audio leaves the machine.
02 / Authority
Fresh utterance per high-impact action
Sessions stale. Resumption of a high-impact tool requires a fresh confirmation phrase that names the action. Patent-protected: voice-gated deterministic tool invocation. A compromised session token cannot escalate without the operator's live voice.
03 / Recognition
Whisper-class STT on-device
On-device speech-to-text with multi-language coverage and real-time streaming transcription. RNNoise cancellation pipeline keeps recognition reliable in noisy environments. Custom wake-word engine with configurable trigger phrases and false-positive filtering.
04 / Synthesis
Piper TTS on-device
Neural text-to-speech with multiple voice profiles, emotion injection, and SSML control. No cloud TTS, no third-party voice rendering, no audio fingerprint shipped off the machine.
05 / Watermark
AudioSeal dual-layer watermark
Every Jarvis-rendered utterance carries a spread-spectrum waveform signal plus an ML-DSA-65 cryptographic seal in the metadata. The mark survives compression and re-encoding, so a downstream listener can tell synthesised Jarvis output apart from cloned audio.
06 / Extreme
Extreme-environment recognition
Jarvis handles cold-induced vocal-tract changes, pressurised cabin acoustics, and EVA-suit helmet-and-mic distortions. Maritime, arctic, aerospace, defence: speaker verification holds where commodity systems fail.
07 / Gating
Voice-gated tool invocation
When Sentinel asks for fresh authority, Jarvis prompts the operator, captures the utterance, runs verification, and signs the gate envelope into the OAR substrate. The signed envelope carries the speaker hash, the action identifier, and the timestamp.
Patent anchors
Jarvis sits on three of the 31 filed UK patent applications behind the Mickai SIOS. Patent 13 anchors voice-gated deterministic tool invocation; patent 06 covers extreme-environment speaker verification; patent 11 covers AudioSeal dual-layer watermarking on synthesised output.
- 13Voice-Gated Deterministic Tool Invocation, high-impact tools require a fresh utterance match at call time.
- 06Voice-Biometric Extreme-Environment Verification, speaker match in maritime, arctic, and aerospace operations.
- 11AudioSeal Dual-Layer Watermark, waveform signal plus ML-DSA-65 seal in metadata, compression-survivable.
UK00004373277 · 31 filed UK patent applications · 914 claims
Wired with
- On-device speech recognition (Whisper-class STT, multi-language)
- Hardware-bound voiceprint in the secure enclave
- Voice-biometric speaker verification (no audio off-machine)
- Fresh-utterance authority for high-impact actions
- Piper neural TTS, multi-voice, SSML, emotion injection
- AudioSeal dual-layer watermarking on every rendered utterance
- RNNoise cancellation pipeline for noisy environments
- Custom wake-word engine with false-positive filtering
Read the Jarvis patents.
The Jarvis voice authority is filed as part of 31 UK patent applications behind the Mickai SIOS. Read the claim text on the patent surface, or download Mickai and enrol your voiceprint on a sovereign host yourself.
Engineered by Micky Irons in Cumbria, United Kingdom · @mickyirons