Are cloud-based voice assistants doomed? That seems like an overly dramatic question considering the current status of millions of users of Google Assistant, Amazon Alexa, and Apple’s Siri, but we’re not so sure about the future. Google and Amazon recently pulled out of their voice assistants, with Amazon laying off much of the Alexa team as it lost $10 billion a year. Google isn’t quite at the “fire everyone” stage just yet, but it’s reportedly less interested in supporting the Assistant on third-party devices, which would be a crippling move given Google’s extremely small hardware division. Everyone built these systems under the assumption that there would be a revenue stream later, but that revenue never came, and it’s starting to look like the bubble is about to burst.
A project that relies heavily on Big Tech voice assistants is not waiting for doomsday. The Home Assistant team declares 2023 “Home Assistant’s year of Voice”. This is actually the leading smart home project that says, “If these cloud-based voice assistants don’t bring Big Tech a multi-billion dollar revenue stream, that’s fine, we’ll do it ourselves!” There are already a few burgeoning, open source voice assistant projects, but the Home Assistant team has proven it can handle a big project. It has a huge, thriving community and enough revenue to support full-time employees, making it the new front-runner for a viable local voice service.
Plus Home Assistant isn’t starting from scratch — it went after what it called the “most promising” open source voice assistant out there, “Rhaspy,” and hired the lead developer, Mike Hansen, to work full-time on voice in Home Assistant . Hasen will now join Nabu Casa, the Home Assistant commercialization company. According to Home Assistant founder, Paulus Schoutsen, “Rhasspy stands out from other open source speech projects because Mike doesn’t just focus on English. Instead, his goal is to make it work for everyone. This is going great since Rhasspy has been around for 16 different languages today.” The plan is to support all 62 languages that the Home Assistant currently supports, but with voice, all without the need for an internet connection.
Schoutsen says Home Assistant will keep the scope of the project “manageable” for now and will “limit the number of possible actions and focus on the basics of interacting with your smart home. No web searches, calls or voice games.” there’s probably a way for the Home Assistant community to add extra features to it, and then it will quickly grow into imitating Hal 9000 and doing a million other extra things.
The problem with all these Big Tech voice services is that they have no way of generating ongoing revenue. They don’t really have a way to serve ads and nobody wants another subscription service. However, they generate ongoing costs due to the server time required to process all that voice communication. Google and Amazon exacerbated the problem by selling their speech hardware at cost in an attempt to win the voice assistant rush, hoping to add an additional revenue stream later on. Apple launched a high-end Siri speaker, the HomePod, in 2018 for a then shocking $350 price tag, but in hindsight that looks a lot more durable than whatever Amazon and Google did.
Google seemingly tried to solve this problem with its second-generation speech speakers, which moved “some” speech processing to local chips. Moving some of the speech processing out of the cloud reduces server time, but it’s unclear if that’s enough to meet Google’s incredibly high standards for ongoing product support.
By moving a voice assistant to a host-it-yourself environment, you can afford those costs at a reasonable rate (some kind of hardware to start and then just the electricity bill) without worrying about the weird, shifting priorities of a Big Tech ecosystem. There will probably be a lot of other benefits as well. A local voice assistant would be great for privacy, and it’ll feel less like having an always-listening internet tap in your house 24/7. Home Assistant has also proven that keeping all of these local results provides a much faster experience for your smart home, and Google has said the same about its limited local speech processing.
The Home Assistant already has a text-only “Conversation” command system, so it’s a matter of building that out and hooking it up to voice input and output. What’s unclear about this is the hardware solution beyond the usual “pile of wires and circuit boards” that usually dominate open source smart home projects. Part of what makes the Google Assistant and Alexa so popular is a selection of nice speaker/mic combinations that you can scatter around the house without looking like a mad scientist. The Home Assistant commercialization company, Nabu Casa, has been working lately to close this hardware gap, creating plug-and-play server boxes and dongles to bring things like Matter support into the Home Assistant ecosystem. Maybe it can launch a loudspeaker.
