Picking Locks from Audio

That is picking locks of the physical, notched metal kind.  Seems a machine learning kind of approach would be possible. Learn from the sounds emitted when opening or manipulating a lock.  The author seems skeptical. Further the NUS team looks at what crimes might be possible with emergent tech, leveraging sensing, networks and machine learning.  A good follow.

Picking Locks with Audio Technology
By Paul Marks  ACM
August 13, 2020

The next time you unlock your front door, it might be worth trying to insert your key as quietly as possible; researchers have discovered that the sound of your key being inserted into the lock gives attackers all they need to make a working copy of your front door key.

It sounds unlikely, but security researchers say they have proven that the series of audible, metallic clicks made as a key penetrates a lock can now be deciphered by signal processing software to reveal the precise shape of the sequence of ridges on the key's shaft. Knowing this (the actual cut of your key), a working copy of it can then be three-dimensionally (3D) printed.

This discovery of a major vulnerability in the physical keys that millions of us use to secure domestic and workplace doors and lockers was made by cyberphysical systems researcher Soundarya Ramesh and her team at the National University of Singapore. At the 21st International Workshop on Mobile Computing Systems and Applications (HotMobile 2020) in Austin, TX, in early March, Ramesh revealed how their technique works.

What's being attacked by the NUS team are the keys to pin-tumbler locks, best known as Yale or Schlage keys, though those are just the market leaders and a whole host of other firms make them, too. Inside such locks, six metal pins, affixed to springs, are pushed up to different heights by the ridged teeth on the key, or kept low by the voids between the ridges. When all six spring-loaded pins are pushed to the correct height by the right key, the tumbler containing them is freed to turn, allowing the lock to be opened. Such a lock typically has something of the order of 330,000 possible key shapes.  ... " 

General approach: " ... The NUS team, which studies sensing and embedded and network security, previously investigated potential future crimes that tech might allow to happen. Last year, for instance, they developed a way of fingerprinting the sound of parcel courier drones, to distinguish them from criminal attack drones that might impersonate the real thing to steal valuable parcels awaiting pickup.  ... "?