Nanoveu’s EMASS has demonstrated the first-ever fusion of real-time keyword detection and voice recognition running simultaneously on its ECS-DoT edge AI co-processor at sub-milliwatt power, delivering fast, private voice intelligence with minimal battery impact.
- First dual AI function fusion on ECS-DoT chip
- Keyword detection achieves 92% accuracy with 4ms latency
- Always-on operation at under 0.5 milliwatts power
- Audio processing entirely on-device for privacy
- Platform supports multimodal sensing beyond voice
Breakthrough in Low-Power Voice AI Fusion
Nanoveu Limited (ASX:NVU) has unveiled a significant milestone through its EMASS unit, showcasing the simultaneous operation of keyword detection and voice recognition on a single ECS-DoT edge AI co-processor chip. This fusion runs entirely on-device at a remarkably low power envelope of 400–500 microwatts, or less than half a milliwatt, without waking the host processor until a command or authorised user is detected. The demonstration marks the first time ECS-DoT has handled two AI functions concurrently, delivering near-instantaneous voice intelligence with a latency of approximately 4 milliseconds and a keyword detection accuracy of 92%.
Addressing the Power-Latency-Privacy Trade-Off
Voice interfaces are rapidly becoming the primary mode of interaction with computing devices such as tablets and PCs, but always-on voice processing has traditionally struggled with balancing power consumption, responsiveness, and privacy. Conventional architectures either keep the main processor awake; draining battery life; or duty-cycle it, which introduces lag and risks missed activations. Nanoveu’s ECS-DoT chip sidesteps these challenges by performing continuous keyword spotting and voice recognition on a dedicated co-processor at sub-milliwatt power, waking the host CPU only when necessary. Crucially, all audio processing happens on-device, meaning audio data never leaves the chip, significantly reducing privacy risks associated with cloud-based voice assistants.
Robust Performance on Standard Hardware
The dual AI functions were implemented on an ECS-DoT evaluation board using a standard PCM digital MEMS microphone, requiring no specialised hardware. The keyword detection model, a compact 8-bit neural network, was tested over 1,500 times, achieving a top-1 accuracy of 92% and a top-3 accuracy of 97%. The voice recognition model generates a per-user voice signature to authenticate authorised users, enabling personalised and secure device responses. This combination allows devices to respond instantly and gate interactions appropriately, all while maintaining minimal power draw.
A Platform for Multimodal Intelligence
Beyond voice, ECS-DoT’s architecture supports multimodal sensing, capable of running image classification and motion sensor models alongside audio, all within the same ultra-low-power budget. This positions ECS-DoT as a versatile always-on co-processor for next-generation devices, enabling layered sensor fusion without the power and thermal costs of keeping the main processor awake. Nanoveu’s CEO of Semiconductor Technologies, Mark Goranson, highlighted this versatility, noting that the platform offers device makers optionality to add low-power intelligence across voice, vision, and other sensors.
Commercialisation and Next Steps
EMASS is advancing the technology toward commercial deployment by expanding the keyword vocabulary, improving robustness across diverse users and environments, and developing production-ready reference designs for computing devices. The company is actively engaging with manufacturers, processor vendors, and platform integrators to integrate ECS-DoT-based voice processors into upcoming tablets, PCs, and always-on consumer and industrial endpoints. This development complements Nanoveu’s broader edge AI strategy, which recently demonstrated notable energy efficiency gains in drone applications, reinforcing the chip’s versatility across markets. Nanoveu’s ECS-DoT AI chip has already shown promise in other low-power AI tasks, bolstering confidence in its commercial potential.
Bottom Line?
Nanoveu’s ECS-DoT chip sets a new benchmark for low-power, privacy-conscious voice AI, with its multimodal capabilities hinting at broader applications that could reshape always-on device intelligence.
Questions in the middle?
- How quickly will Nanoveu secure design-in contracts with major device manufacturers?
- Can ECS-DoT’s multimodal sensing capabilities extend effectively beyond voice to vision and motion in commercial products?
- What competitive pressures might emerge as other semiconductor firms pursue similar low-power edge AI solutions?