Welcome to the first in our series of blog posts featuring papers from the recent Acoustics in the Sun conference. Today we’re giving our take on Alex Foster and Deb James’s paper titled AI in Architectural Acoustics.

Is AI just another trend?

There were a lot of AI papers presented at the Acoustics conference, which is understandable. AI is a hot topic, so why not AI in acoustics too? I admit my imagination immediately went to robots being trained to do the onsite visits. How handy would that be? But the reality is that AI is most useful for paperwork (much like in every other industry). Or, in the words of Foster in James, “the automated analysis of architectural drawings to determine key project criteria”.

Classification is king

If you’ve ever used AI to create an image, you’ll be aware of some unexpected and hilarious results. What about architectural drawings? Can the AI map and classify rooms, and specify noise levels?

In short, the authors concluded that it can (if well trained). This is an incredible boon for acoustic consultants, who can potentially reduce the time consuming process of analysing architectural drawings. It can also improve consistency, especially when dealing with large datasets.

What are the limitations?

The main benefits of AI comes with efficiency. A well-trained AI can accurately and consistently remove some of the more labour-intensive tasks. But are there limitations?

You’ll be aware that here at Soundmask we are all about the human touch. What is great in theory, can slowly become an acoustic nightmare for the end users. Thankfully, the research paper concludes that the use of AI is no substitute for a qualified acoustic consultant, and seems unlikely to replace them any time soon. Much like AI in other areas, it is a tool for humans, not an alternative to them.

Conclusion

AI is useful to extract information and classify it. Its accuracy is very depended upon the quality of training.

If you’d like to delve deeper into the world of AI in architectural design, you can find the full paper here.