Scientists improved AI-based speech recognition for home appliances

Many different devices already have some voice recognition software. For example, this kind of technology helps you communicate with your car and your phone. But what about your home appliances?

Scientists at the University of Waterloo and AI company DarwinAI have developed a new AI-based speech recognition technology, which can be tailored for specific tasks.

New AI-based speech recognition does everything locally, but also fits on a small cheap chip. Image credit: Harland Quarrington via Wikimedia (OGL v1.0)

There are already voice assistants that are trying to do everything. For example, Amazon Echo and Google Home are deep-learning artificial intelligence technologies, able to control various smart home appliances. However, these systems have their disadvantages. For example, they are not very efficient and other devices, like home appliances, would have to be integrated into those systems. But scientists have an alternative.

Scientists created an AI speech recognition software that is so compact it can fit on a chip that is smaller than a postage stamp. This means that AI-based operations wouldn’t have to be performed on the cloud, which would further increase data security and privacy. It would also cost only a few dollars to make. This kind of efficiency is achieved because AI is taught some specific commands, like “yes,” “no,” “on” and “off” and can go on from there, finding the closest match for the words it heard from the user. Of course, this means that the software is only good enough to do one specific job – it is not very universal at all, but it can do one job very well.

The efficiency is boosted by attention condensers – some building blocks that allow the AI algorithm to focus  on the most relevant information in sound waves. And all of this results in a very compact and relatively cheap software that could affordably operate systems in vehicles and devices for people with disabilities. Alexander Wong, one of the creators of this system, said: “Cost and efficiency are two of the biggest bottlenecks to the widespread adoption of machine-learning AI. This technology significantly addresses those issues and enables a new class of voice assistants for everyday devices with energy-efficiency needs.”

Scientists are now working on applying the core of this technology to one stand-alone device or software, which would also understand visual cues and text. This should be a very useful system, which would make AI approachable for all kinds of devices.

Scientists say that this technology would be particularly useful for disabled people. But not just that. It could make some home appliances smarter while not increasing their costs significantly. You pretty much could just shout to turn on your washing machine.


Notify of
Inline Feedbacks
View all comments
Would love your thoughts, please comment.x