Edge Impulse Harnesses Generative AI for Synthetic Data Creation

The integration of Synthetic Data presents a novel and effective approach to utilizing Edge Impulse for LLM-based data creation. This integration enables the use of DALL-E for image generation, Whisper for creating speech elements for keyword spotting, and ElevenLabs for audible events. In addition to these features, enterprise customers can incorporate custom LLM sources, such as other data providers or self-hosted LLMs. Furthermore, additional LLM toolkits will be introduced in the upcoming months. These enhancements complement Edge Impulse's existing direct integration with NVIDIA Omniverse Replicator, a framework designed for developing custom synthetic data generation pipelines to produce highly realistic, physically based datasets tailored for training computer vision models.

Within the new Synthetic Data integration, users can swiftly and effectively add and refine their prompts. The resulting output, which includes images and audio fragments, is promptly displayed for evaluation. This allows users to efficiently assess and fine-tune their prompts until they achieve the desired dataset. The iterative workflow facilitated by this integration simplifies the process of identifying the appropriate prompts for data generation. Moreover, any retained data is automatically integrated into the project, ensuring seamless data management.

This innovative toolset significantly streamlines the process of generating and refining prompts to create the desired dataset. It offers an efficient workflow for constructing models using synthetic data, making it easier for developers to generate high-quality datasets by leveraging generative AI.

Currently available for Enterprise-tier users, this new feature will soon be accessible to Professional Plan users. Positioned as a top tab within the "data acquisition" section of Edge Impulse, alongside Dataset, Data explorer, and Data sources options, this feature provides a user-friendly interface for leveraging synthetic data in model development.

For more information, visit www.edgeimpulse.com.