QuickLogic subsidiary SensiML has added generative AI to enhance Data Studio, its dataset management application for IoT edge devices.

This capability allows embedded device developers to use text-to-speech and AI voice generation to create hyper-realistic synthetic speech datasets for building robust keyword recognition, voice command and speaker identification models.

Using these rapidly generated speech datasets, developers can easily create speech recognition AI models using SensiML’s AutoML development tools. These models are optimised to run autonomously and efficiently on low-power microcontrollers in edge IoT applications.

By leveraging speech generation technology from ElevenLabs (elevenlabs.io), Oregon-based SensiML’s feature simplifies the creation of large high-quality datasets. Developers can generate synthetic speech data with realism and tailored voice attributes such as pitch, cadence and tone to meet application requirements. This eliminates the time-consuming and costly process of manually recording phrases from large populations of diverse speakers, accelerating time to market for voice-enabled IoT devices.

Designed for user friendliness, the TTS and AI voice generation feature enables seamless integration into existing Data Studio workflows. Benefits include:

• High-quality voice output: Produces natural and expressive voice samples, enhancing user experiences
• Versatility: Supports a wide range of languages and dialects, catering to diverse global markets
• Efficiency: Streamlines the process of integrating voice generation into AI models, reducing time to market
• Scalability: Suitable for applications of all sizes, from small IoT devices to large-scale deployments

“With the introduction of this generative AI feature into our Data Studio application, SensiML continues to push the boundaries of what’s possible in AI for IoT,” said Chris Rogers, CEO of SensiML. “Developers can now harness state-of-the-art synthetic speech technology to create highly accurate and diverse training datasets, accelerating the deployment of intelligent voice-controlled applications directly on microcontrollers.”

The created datasets are seamlessly compatible with SensiML’s Analytics Studio and its open-source AutoML tool, Piccolo AI, facilitating a smooth transition from dataset creation to model deployment.

For a real-world example, consider a smart-home security system that uses voice commands for activation and status updates. With SensiML’s text-to-speech and AI voice generator (sensiml.com/blog/data-studio-t2s-and-voice-generation), developers can efficiently create extensive voice datasets, enabling the system to recognise a wide range of user commands accurately. This accelerates the development and deployment of the system, ensuring homeowners benefit from reliable and responsive security without the need for constant internet connectivity.

This marks an advancement in the capability of developers to custom-build their own ML code for IoT devices needing to handle complex voice and sound recognition tasks directly on-device, without the need for constant connectivity or high computational power. It is particularly beneficial for applications in environments where connectivity may be inconsistent, and where fast, reliable processing is crucial.

SensiML (sensiml.com), a subsidiary of QuickLogic (www.quicklogic.com), provides software that enables low-power IoT endpoints that implement AI to transform raw sensor data into meaningful insights at the device itself.