Remotely measuring body size, audio documents & congestion management systems

---

1. Shopify – measuring body size

Correctly identifying peoples’ body sizes and making sure that the clothes that people order online fit is a big problem. The inability to correctly judge whether clothes ordered online will fit is bad for business in two ways: 1) some people are less likely to make a purchase online; 2) badly fitted clothes create huge costs with returns.

In this new filing, Shopify also reveal they’re looking into this space.

Shopify describe determining a person’s body measurements using images or 3D data captured in augmented reality. Essentially, Shopify will use object detection to identify a person’s body and then use the body’s distance from the camera to estimate a person’s body measurements. Moreover, Shopify is exploring showing the clothes a user is interested in with augmented reality. With this, potential customers could virtually try on the clothing to make sure it fits well.`

Obviously, clothing is a complicated thing. There are different materials that are supposed to be worn in different ways. Some fabrics are supposed to be worn tightly hug the body, other fabrics like skirts are tight in some areas and loose in others. So in the filing, Shopify describe exploring the behaviour of different materials and adjusting the body measurements to ensure that the clothing fits in the way it’s intended.

The combination of augmented reality, machine learning and a hyper-competitive e-commerce space, will push one company to nail the problem of ordering correctly fitted clothing online. In fact, billionaire Marc Lore (founder of jet.com) recently discussed this problem as being one to solve if you wanted to have a $100m exit in a few years – take a listen.

2. Microsoft – generating audio documents

No pretty picture for this one.

If you read this issue of Patent Drop, you would know about how Facebook is looking at building ‘voice avatars’ that will read out messages in the voice of the sender to the recipient.

This latest filing application from Microsoft is looking to do something similar, except with documents.

Currently, the automated approach of turning texts into audio usually results in an audio file that reads out the text in a single tone. The audio doesn’t differentiate for different characters or the context of a document.

Microsoft is looking to generate different voices for the different roles within a document. For example, consider this sentence: Tom said, “it’s beautiful here”. Microsoft want their text-to-speech to have a different voice for the descriptive text – i.e. Tom said – and another voice for the specific utterance from Tom – “it’s beautiful here”.

The applications of this kind of technology are pretty wide. To take one, Microsoft mention the rise of audio stories. Most audio stories are read out by just one human voice actor – something that takes time, is expensive, and still limited in the ability to tell a story with different voices for different characters.

Another startup called sonantic.io is working on this problem with the initial focus on voice acting in games.

One could also imagine emails being read out in the voice of the sender, as a way for people to catch-up on emails in a more human, personalised way.

With audio being a hot space when it comes to connection (i.e. Clubhouse) and media consumption (e.g. Podcasts), it’s interesting seeing AI being applied to generating synthetic voices and the various contexts where this could be useful.

3. Google – reducing vehicular congestion

This is a fun one.

When vehicles go through intersections and are coming up against congestion, drivers will rely on their experience to decide when and how fast they should accelerate. Moreover, the information that each driver has is limited to the car they see in front of them and the car they see behind them. As a result of this, fewer cars end up passing through an intersection than is possible. In the filing, Google describe that certain traffic light that display green for 90 seconds, allow approximately 300 vehicles to pass through an intersection at a normal speed, but only 200 vehicles may end up passing through the intersection because of inefficient acceleration and deceleration.

In this filing, Google describe a congestion control system where vehicles coordinate their acceleration so that more vehicles can travel through an intersection, reducing journey times and increasing fuel efficiency.

For example, if there were multiple self-driving vehicles at an intersection, the cars would communicate with each-other to synchronise movement in a way that would enable the cars to maximally accelerate while maintaining a safe distance between the vehicles.

However, Google isn’t relying on the whole population to switch over to autonomous cars before this congestion system can be implemented. One potential implementation they’re exploring is delivering instructions to drivers to manually accelerate in optimal ways via navigation systems. In the same way that Waze crowdsources driver information to generate traffic data and in turn navigate drivers to move away from high-traffic routes, Google could look to create a congestion communication system between cars using Maps.

The topic of networked information is a fascinating one. In this instance, Google is looking to have cars communicating with each-other in order to maximise driving efficiency. It’s difficult to dispute that this is for the good. But on the other hand, Amazon’s network of Rings is now creating a citizen surveillance network that is available to police departments. This is a lot more murkier.

Overall, the network of connected devices and machine learning can both yield incredible efficiencies for society, as well as pretty dystopian outcomes. Welcome to the future, where everything is up for grabs.