April 13, 2018
Hollywood has confused our understanding of sounds
TechRadar came to talk to us recently about our cutting edge AI sound recognition software and its potential across consumer products.
In the resulting coverage, the actor-turned-journalist, Andrew London took an interesting angle on the subject, referring to our discussion of Foley. When we get interviewed by the media it is often a subject that we touch on. Most people’s understanding of sound comes from what they’ve heard on TV, especially when it comes to a window being broken. Take a pillow case and fill it with small metal objects such as screws, nut and bolts, then add small glass vials some filled with water and some without and now bang the filled pillow case on a surface. There you have your window glass break; a huge amount of effort is spent making the sound of something sound more movie-like.
In the article, TechRadar uses a great example in a video asking readers to decide whether the sound they hear is rain or bacon:
In the real world sound is very different and when you’re creating a whole new AI capability around the sense of hearing, you have to really understand it in great detail. If you’d taught the technology by feeding it movie clips then it wouldn’t work if your window was smashed for real, but it would every time the Hulk throws somebody through a window. As a result, this means collecting those real world sounds in real environments and then building the world’s largest database of real world sounds (which we’ve done).
As Chris said to Andrew during the interview:
“"We have the sound lab in Cambridge (UK), where we smash, bash, crash, and beep our way through a whole range of different sounds to explore how the ideophones change with different variations of the sound."”
When we show off our technology we often find ourselves explaining the important distinction between real-world and either fake or poorly recorded sound. We are giving machines a greater sense of hearing and it is important that these products (like the recently released Hive Hub 360) understand what things really sound like and not what we think they sound like.
That won’t fly with consumers. They want accuracy, precision and reliability, especially if we are promising them a smart product that will help them out. However, if you’re looking for a Hulk detector…