Overview
This webinar, hosted by EETech and All About Circuits, focused on enabling voice user interfaces (VUIs) on microcontrollers using Cyberon's dSpotter solution and Renesas MCUs. Speakers discussed market trends, technical challenges, solution features, and practical implementation, followed by a live Q&A.
Market Trends and Challenges
- Global VUI market is projected to reach $23B by 2025, driven by smart speakers and IoT adoption.
- COVID-19 has accelerated the need for voice-based, touch-free interfaces.
- Voice technology segmentation: speech recognition (commands, local processing) vs. conversational AI (natural language, cloud-based).
- Key design challenges include limited MCU resources, need for high accuracy in noisy environments, and the burden of data collection/training.
Cyberon dSpotter Solution Overview
- dSpotter is a local, always-on high-accuracy voice recognition engine for embedded devices.
- Operates fully offline—no network, low latency, and strong privacy.
- Uses a phoneme-based model; command customization is fast, requiring only text input.
- Supports over 40 languages including specific regional versions and bilingual models.
- Optimized for low resource consumption (≈40 MHz, 200KB flash, 50KB RAM on Cortex M4).
Development Workflow and Tools
- The dSpotter Modeling Tool (DSMT) allows developers to define, test, and optimize custom commands without collecting training audio.
- Online and offline testing functions verify recognition and help tune performance.
- Integration with Renesas MCUs is straightforward: model generation, importing to MCU, coding actions, and device testing.
- Supports additional software features: voice activity detection (VAD) for power savings and compressed audio playback for voice responses.
Performance and Demonstration
- Demo showed real-time wake word/command recognition, robustness to background noise, and ease of multi-language support.
- Benchmarking exceeded accuracy targets (often >95%) under varied noise levels and distances.
Licensing, Support, and Next Steps
- dSpotter is a software-only solution requiring a single microphone input.
- Renesas provides license-free access to dSpotter and the DSMT tool for development—contact Renesas to apply for access.
- Supports up to 100 commands on typical MCUs; bilingual and multi-language modes depend on MCU resources.
Q&A Highlights
- Only software components required; hardware is standard MCU and microphone.
- License and demo toolchain available via Renesas.
- Typical requirements: 40 MIPS, 40 MHz CPU, one microphone.
- Multi-language and bilingual models supported depending on platform.
- dSpotter is robust to noise, but does not itself perform noise reduction.
- Developers can tune recognition sensitivity using DSMT tool and provided guidelines.
Summary and Call to Action
- Edge-based voice UI is now practical and efficient on MCUs with Cyberon and Renesas solutions.
- Supports diverse languages, is highly portable, and easy to integrate.
- Interested developers should contact Renesas for immediate prototyping access and future updates.