Google introduces LiteRT: a new framework for high-efficiency AI on devices with NPU

Google has introduced LiteRT — an innovative cross-platform framework designed to radically enhance the efficiency and speed of artificial intelligence (AI) directly on user devices. This tool is specifically developed to maximize the potential of neural processing units (NPUs) integrated into modern gadgets, from mobile phones to Internet of Things solutions. The main objective of LiteRT is to provide developers with a powerful platform for implementing and optimizing complex AI models, ensuring their seamless real-time operation.

Integration with NPUs is a key element of LiteRT's architecture, allowing the framework to perform resource-intensive AI computations with unprecedented speed and minimal power consumption. Traditionally, complex machine learning algorithms required significant computational power, often only available in cloud data centers. LiteRT changes this approach by optimizing the execution of such tasks directly on the device. This not only significantly reduces latency and improves the responsiveness of AI systems but also extends device battery life, making AI more accessible and practical for everyday use.

LiteRT's capabilities find application in a wide range of tasks requiring high real-time AI performance. These include the implementation of advanced video effects, such as intelligent background blur, image quality enhancement, or the application of AR filters in video chats and streaming services. The framework also significantly improves the performance of speech recognition systems, making them faster and more accurate, as well as motion capture technologies, which are critical for interactive games, virtual reality, and augmented reality. A striking example of its practical benefit is the potential to enhance the functionality of applications like Google Meet, where users will be able to enjoy smoother and more sophisticated AI features without a noticeable impact on device performance.