Funding: Data2Sustain EDIH
Ideal Technologies (DMAC Technologies Ltd.)
Ideal Technology specialises in assistive technology and environmental control systems for people with disabilities. Based in Dublin, they provide voice-controlled home automation for users who cannot physically operate conventional switches, controls, or emergency call devices.
Unlike commercial smart home systems built for convenience, VoiceMate addresses life-critical needs. Cloud-based voice assistants fail during internet outages – common in rural Ireland- with consequences extending beyond inconvenience to genuine safety concerns. For a wheelchair user without broadband or an elderly stroke survivor unable to reach controls during network failure, offline reliability is a fundamental requirement, not a preference.
Having validated microphone hardware through an earlier Innovation Voucher with WiSAR, Ideal Technology faced a pivotal choice: legacy grammar-based SAPI 5 (stable, documented) versus modern neural-network ONSE SDK (promising but untested for their use case). This architectural decision required systematic evaluation.
The Challenge
Selecting the wrong speech recognition engine would require costly rework after significant development investment, potentially delaying market entry. Published benchmarks focused on cloud services under ideal conditions, not representative of real assistive environments with ambient noise, variable microphone positioning, and diverse accents.
Critical requirements included: fast, first-time-accurate recognition for users who cannot easily repeat commands; robust performance under realistic noise (television, conversations, appliances); and reliable handling of Ireland’s multicultural care sector accents. Could modern offline neural recognition deliver sufficient advantage over established grammar-based systems to justify the transition?
WiSAR Solution
WiSAR Lab designed and executed a rigorous Test Before Invest evaluation spanning controlled laboratory testing, realistic noise simulation, and public demonstration validation.
Phase 1 – Parallel Test Application Development
Engineers developed two independent Windows Presentation Foundation (WPF) applications with identical interfaces, test datasets, and data logging, differing only in recognition engine. The SAPI 5 application used grammar-based recognition; the ONSE application employed ONNX neural models operating entirely offline. Both recorded every attempt to an SQLite database, capturing recognised text, confidence, latency, and word error rate.
Testing took place in different environments – a quiet anechoic chamber location and simulated noisy office environments (80–90 dB(A)), wired and wireless microphones, six speaker accents, and both Assisted Living commands and phonetic test sets. From 189 test runs, 143 were validated after quality filtering, generating robust statistical evidence.
Phase 2 – Demonstration and Market Validation
WiSAR developed a VoiceMate Launcher demonstrating ONSE in an interactive smart-home simulation, integrating wake-word detection, structured command parsing, and natural speech synthesis, all offline. Deployed on a GMK i5 mini-PC in a custom 3D-printed housing, the system operated continuously at Rehacare 2025 in Düsseldorf without internet, validating laboratory findings and serving as a networking platform.
Deliverables included comprehensive technical documentation, working proof-of-concept, structured performance databases, evidence-based recommendation, and professional demonstration hardware. The Rehacare exhibition provided ecosystem networking, connecting Ideal Technology with European distributors, care providers, and industry partners.
Impact & Benefits
- Quantified Performance Advantage: The ONSE outperformed SAPI 5 across every metric: 18 percentage point accuracy improvement (76% vs 54% in noise), 14 percentage point word error reduction, 220 millisecond latency improvement, and strong confidence-accuracy correlation (r ≈ 0.87).
- De-Risked Major Investment: The evaluation provided evidence-based confidence to commit resources. Ideal Technology gained quantified performance data specific to their use case, dramatically reducing technical and commercial risk.
- Validated Resilience: Testing confirmed superior performance under realistic noise and across diverse accents – critical for multicultural care environments.
- Market Development: Rehacare 2025 generated interest from European distributors and care providers, opening wider market potential through networking that would be difficult to achieve independently.
- Accelerated Development: Test Before Invest compressed 6-12 months of potential internal trial-and-error into a focused 4-month engagement.
