OpenAI is ramping up efforts to upgrade its audio artificial intelligence models as it prepares to launch its first AI-powered personal hardware device, according to people familiar with the matter.

Three people with knowledge of the project said the device is expected to center on audio-based interaction, making voice the primary interface between users and the system.

While users can already converse with ChatGPT via voice, the large language model that supports its audio capabilities is not the same as the one powering its text-based interactions, the sources said. OpenAI researchers believe the current audio models lag behind text models in both response accuracy and latency.

To address these gaps, OpenAI has in the past two months consolidated multiple engineering, product and research teams to focus on improving audio model performance, the sources said, so that the technology can meet the demands of a future consumer hardware product.