Open source AI-powered voice assistants such as Rhasspy, Mycroft and Jasper have aimed to provide privacy-preserving, offline experiences without compromising functionality. However, the development has been slow due to the inherent challenges of creating an assistant and the extensive R&D and infrastructure of tech giants like Google Assistant, Siri, and Alexa.
The German nonprofit, Large-scale Artificial Intelligence Open Network (LAION), has taken on the ambitious initiative of building a fully open voice assistant, named BUD-E, that can run on consumer hardware. This new project aims to leverage emerging GenAI technologies and large language models (LLMs) to create a natural and engaging voice assistant.
BUD-E is a collaborative effort between LAION, Ellis Institute, tech consultancy Collabora, and the Tübingen AI Center. The roadmap for BUD-E includes enhancing emotional intelligence, handling conversations with multiple speakers, reducing hardware requirements, and minimizing latency.
While BUD-E is available for download on GitHub, it is still in its early stages. LAION has patched together several open models to create an MVP and is working with Collabora to customize open source speech recognition and text-to-speech models for BUD-E.
The team acknowledges the need for a well-functioning natural voice assistant and is committed to making BUD-E accessible to a wider audience. While addressing issues of accessibility and potential biases, LAION remains dedicated to transparency and ethical guidelines in the development of BUD-E.
