When we talk about AI and datasets, it's usually images. Or text. But what about sound? Doesn’t every living thing or soul make sound? If sound is everywhere, we also have data everywhere! But what sound can we capture, and how to find use for it? To give a few examples, here's what AI + Audio could do to reduce cost, increase quality, improve communication, speed up processes, and make living safe & sound!
02/ Operational excellence
As a first example, manufacturing is an area that holds great potential for operational excellence! With AI, you could track the sound of an operating machine, to better optimize your production and get better timing for changing tools. You could also use AI for early warnings by recording the operating sound of a machine, and get alerts before something (seemingly unpredicted) really happens.
Can we record operating sound to get early warnings with AI?
03/ Connectivity
As our homes, places, and spaces get increasingly connected, we also get more dependent on hardware. But what if we could use mini microphones instead of sensors; to cut cost, maintenance and installation time? With sound (or no sound) everywhere, we have a chance to record everyday audio with a tiny microphone instead of tons of sensors. With AI, we can locate odd sounds among the normal - and give us a chance to automatically switch off the lights when a room is silent and empty, or stop a thief when a window breaks with a crack, or to roll in a sun visor when the stormy winds start to whistle.
To make it easier to work with sound, it's a good trick to translate waves into images, like we did with our customer Epidemic Sound (continue reading...). Once you have a dataset with images it's pretty easy to use classification models to apply on your sound task. For you who are curious and want to explore things at home connected to sound and tracking, here are a few interesting public datasets with everyday sound to play around with; AudioSet and FSD, and this is a smart thing to monitor your home with!
04/ Music
Licensing of music is a fast-growing business as creating content becomes increasingly important. So we teamed up with Epidemic Sound, a Stockholm based music company, to help them categorize music. For two reasons; to speed up the process of sorting songs, and to increase quality by reducing the impact of personal judgment.
By using soundwaves from pieces of songs, we could translate them into spectograms (=images), and once turned into images we could use well-known image classifiers to tag them with different moods. Moods like “Funny”, “Laidback”, “Epic” (and 43 other ones…). It turned out that once we had the preprocessing of data done, and image data ready, it was really quite simple to create models to place music into different categories. If you’re curious about more details, check out the full story about the Epidemic case here.
05/ Environment
Datasets with sounds from animals is something that enthusiasts and environmental specialists have been recording for ages, and with a bit of AI help we could take care of that data and keep track of rare species to protect them, see patterns, and take better care of our environment. Here’s a worrying long list of threatened species from World Wildlife Fund (WWF) of threatened species, and deforestation explained by National Geographic.
06/ Safety
And last but not least - what if we could use AI to detect accidents in traffic? With the help of acoustic sensors and AI classification of sound, we could connect the sound from a car crash to getting an ambulance to help. Actually, Washington D.C. is already doing something similar with their ShotSpotter system. By recording sounds in the streets, the city is keeping track of gunfire (as opposed to fireworks or other similar or distracting sound) to protect its citizens and get a better chance to react faster when accidents occur.
Can we use AI and audio to act faster in emergies?
