How to keep your voice private

Using a voice assistant that runs in the cloud exposes you to massive Privacy abuses. Here is how to prevent it.

This article was originally posted in French in Les Echos

Privacy is back at the forefront. From the upcoming European GDPR regulation on data Privacy to the Cambridge Analytica scandal, it seems clear that both consumer and companies are now taking Privacy seriously.

One particular product category where Privacy concerns keep growing is smart speakers / voice assistants type products such as the Google Home and Amazon Echo. Indeed, now that almost 50M Americans have a microphone connected to the internet in their homes, mass manipulation, surveillance and hacking become real issues to deal with. Just asking users to “trust” a provider is not enough. Voice companies should guarantee Privacy by Design.

My belief is that just like messaging apps before, the perception of Privacy in voice assistants will go from inexistent to absolutely necessary. And it will happen in the next 12 months.

The dichotomy between Cloud and Privacy

But Privacy by Design is hard to achieve, and processing voice in the Cloud makes it virtually impossible. Let’s imagine a best case scenario: a company that processes voice in the cloud, only sending the voice sample that comes after the wakeword (the “Hey Siri” thing), and not storing it after it is processed. But however much they advertize it, they would in fact offer zero privacy.

First, because all the code runs in the cloud, there is no way to know that the open source code is what actually runs. There is really no way to tell if they actually do what they say. Open sourcing is necessary for transparency, but without auditability or the ability to run it yourself, it does not offer any protection.

Second, even though they might not store the voice samples today doesn’t mean nobody can or will at some point. A change in leadership, a rogue employee, a hacker or a government could all decide to spy on you, without you ever knowing about it. Not only could they access your voice samples, they could also turn the microphone on remotely, without any indication of this happening on your side (the LEDs are usually not hardwired to the microphone).

Third, although theoretically only what you say after the wakeword is sent to the cloud, in practice there are a lot of misfires, where the voice assistants lights up thinking you called its name. Whatever you are saying at that moment will be sent to the cloud, possibly without you even noticing it. There could also be bugs that makes the assistant record more than it should, as happened with the Google Home Mini.

Simply put, you would need to blindly trust that company with your Privacy, as you would have no control on it whatsoever. The simple act of sending your voice to the cloud is enough to lose control over your Privacy, and until things like homomorphic encryption start working efficiently, there is simply nothing that can be done about it.

Edge Computing to the rescue

There is however a simple alternative, which is to never access the data in the first place, instead processing the voice of the user directly on the device they are speaking to.

This concept, called “Edge computing” is the opposite of Cloud computing. Instead of centralizing the data and processing it in the cloud, you decentralize it completely by processing it directly on the device, or on the local network it is part of (e.g. your home, office, …).

By doing so, anyone using the product would be able to actually check that no data is being sent, simply by monitoring the traffic going out of the device. They could also turn off updates if they are suspicious, or use the assistant offline, as many use-cases don’t require internet access (think home automation, appliances, cars, etc..). And by making it open source, anyone would then be able to compile and run it themselves, making sure that whatever runs on their device is what is actually advertized.

Since user data would not be centralized anywhere, it would make it exponentially harder for governments to do mass surveillance or for hackers to steal massive amounts of data, as they would need to access each device individually (vs just one company’s servers). Rogue employees would not be able to spy on someone without accessing their device either.

Even bugs would have no impact on your Privacy. The device could record you 24/7, or it could mistrigger hundreds of times a day, it would make no difference since your data would never leave the device anyway.

You would not need to blindly trust an Edge computing company, because your Privacy would be guaranteed by design. And since you would be able to act on it if you were suspicious, you would actually be in control.

You do not need the cloud

The technology to process voice on the Edge already exists, and some of it is even open source. In recent benchmarks, it is shown to perform on par or better than cloud-based solutions, while using vastly less computing resources. It can even run on cheap commodity CPUs such as a Raspberry Pi.

For most use cases of voice assistants, there is essentially no reason to process voice in the cloud anymore. As such, a company that pretends to offer Privacy while processing your voice the cloud is at best clueless, and at worse lying. The technology to decentralize processing of voice on the Edge exists, and is available for anyone to use.

Let’s not risk another massive data breach; our homes are our sanctuaries, and we should keep them Private!

If you liked this article and care about Privacy, smash that clap button, then tweet everyone 👉👉 randhindi & snips

Interested in understanding what’s going on in tech these days? Follow my instagram stories @randhindi 🤓

It is also highly likely we have a job for you at Snips 🤩! We are the largest voice startup in Europe, and are hiring in machine learning, software engineering, blockchain, sales, product, marketing, etc…