View profile

Hearing Voices - Issue #78


Hearing Voices

July 12 · Issue #78 · View online

This week's news about voice computing apps, delivered directly to your inbox.

“How do you discover new skills for your smart speaker? ‘Other’ says 7% of respondents.”
– NPR Smart Audio Report
“How do you discover new skills for your smart speaker? ‘’ says somewhere between 0% and 7% of respondents.”
– Hearing Voices, The ‘Other’ Way To Discover New Skills For Your Smart Speaker™
Did someone forward you this email? Subscribe at

What's Up In Voice
> Audio 3.0 Chat: People are starting to float product ideas that fall into the third wave of audio that I laid out in the last newsletter. Here’s an idea shared on twitter for a product called Stations that would be an Audio Twitch / New Radio. This reminds me a bit of Unmute and somewhat of its evolution into Beep.
> Scan Systems Win at Voice Marketing Meetup. I spoke with about the third wave of audio and they also shared their perspective on audio at Betaworks.
> Also at the event was Future Ear, who did a writeup focusing on Podcast-to-Podcast advertising.
> Listen to Earios - Podcasting by women, for everyone.
> Check out the Voice Assistant SEO Report for Brands by VoiceBot and Magic & Co. Some highlights are that “A January 2019 survey by Voicebot found that nearly 60% of U.S. adults say they have used voice search and 47% expect to increase usage this year.” However this isn’t shocking given we have seen increasing data around voice search on smartphone and they didn’t break out smartspeakers other than to say they’re a “big contributor.” They do, however, acknowledge that search is so concentrated in Google (and they point out that Amazon Alexa uses Bing’s knowledge graph and Appe Siri uses Google’s). Interestingly, it turns out 46.7% of product searches in the US start w/Amazon (compared w/34.6% for Google). And the report highlights that there are specific dominant searches when it comes to content: “YouTube is the top search engine for video and number two worldwide for all searches” (they acknowledge that it’s owned by Google; the point is that people search different places for different things. However the report breaks out the data in an interesting way: number of search results by device:
I find this so fascinating not for the reason they highlight (“Many people recognize that in a voice-only search interaction, a single audible result is the only decent user experience…So, when we talk about voice search, it is important to recognize that the search experience and results will often differ based on the interface.”) I find it interesting because right now most people are trying to find the initial use cases for voice to be built into their products (e.g., big companies launching Alexa skills). However working backwards from the number of search results you want to offer is an interesting approach.
The other interesting graph from the report shares the most common voice search topics:
Note that this is comparing all respondents to smart speaker owners. So, for example, maps and or directions under-indexes on smart speaker searches and finding recipes over indexes. Neither is surprising.
From a product perspective, what’s potentially most interesting is to overlay the two sets of data. For example, asking a fact-based question should yield one result, which means it would fit into the smart-speaker category in the first graph. And as predicted, it over-indexes with smart-speaker owners. Similarly, asking for a restaurant recommendation presumably requires showing multiple results, which means that it shouldn’t fit onto a smart speaker and it under indexes in the second graph. Again, not surprising. What is surprising to me is that whether something over-indexes on smart speaker search appears to be much more highly correlated with number of results than on use-case (e.g., the location of the device). So for example, finding recipes over indexes 3.7% above the 22.9% (16% over index) which one might expect to be the most pronounced. But Celebrity questions, which likely have a single answer but aren’t particularly related to what room you’re in (which is different from recipes), over-index by roughly the same 3.8% but on a 9.7% base, so 40% increase due to it being on a smart-speaker.
Now this is far from a detailed data analysis, but what I take from this is that lowering friction because of the expectation of getting a useful result (in a single search result) has more impact than lowering the friction of the physical room you’re in (e.g., lowering friction because of your physical context). Said another way, if you don’t expect to get the right recipe result, you’re still probably going to look it up on your phone (i.e., the impact of having a smart speaker isn’t that high). But if you expect the result of asking your smart speaker to be pretty good, it lowers the friction and you’re more likely to use the smart speaker to do it, even if the search has nothing to do with the fact that you’re in your kitchen.
Again – not a data scientist – but as you go through each category in the second graph and ask yourself “do I expect to do this because I’m near a smart-speaker” and then also asking “do I expect the smart speaker to be good at answering this” the former doesn’t seem to impact the numbers as much as the latter question.
So when we design voice skills, we should be thinking – how likely is it that when a person thinks of this question, we’ll be able to provide the right answer. Versus how likely is it that when we *think* someone is near the speaker, they’ll ask the question. Subtle difference, but an interesting one, and one that surprised me.
> Also read the NPR Smart Audio Report. This report asked people whether “the screen on your smart speaker has made it easier to discover new content” and also whether it makes the speaker easier to use. Results were 66% agreeing with each statement. I wish they’d asked whether it make them like the device as much (is it annoying to have ads in the background of your Alexa Show while you’re not using it).
Think "Other" should actually be labeled "heard about it on"?
Think "Other" should actually be labeled "heard about it on"?
Top ten weekly smart speaker requests are below and to me, the biggest news is that there is no new news: it’s basically used for basic built-in utilities (weather, alarms, timers), music, and news. Basically none of these top use cases are from specific skills or require any meaningful personalization or search:
Another interesting stat in terms of behaviors is that 58% of those who own smart speakers worry hackers could be using it to gain access to their home or personal information, but 54% trust companies that make smart speakers to keep their personal information secure. I wonder if it’s the same 50%…
Maybe the most interesting thing to me was that of people who don’t own a smart speaker, 43% are not interested in owning one. Only 11% of those who don’t own are very interested. They do break the data down into age groups, and clearly the 55+ set are much less likely to be interested, but to me, at this point, it’s got to be a question of who’s very interested, not just who says they’re interested. To date, we’ve seen massive smart speaker sales, but it begs the question: we know people who already have are buying, but are we at a plateau of new customers until we get some good use cases?
The answer is given partially in that “60% of people who don’t own smart speakers agree they don’t need that kind of technology in their lives.” I’d bet that the reality is that once there is a compelling use case, they might adopt, but until then, they’re not going to. People bought the iPhone not because it was a “smart phone” but because it brought you the “real” web (not that WAP that blackberries used to use). That’s conjecture but I take it for the general direction of the point.
The counter argument to all of this is that the third most cited reason for not having the smartphone is that voice enabled speakers are too expensive. In 2017, that was 60% of people, but in 2019 it’s still 53%. So despite the prices dropping, price is still a point of friction.
Finally, since I’m an Alexa skill developer and many of you are as well, I’ll leave you with this graph:
I really wish they’d called out which were the specific skills. What’s interesting to me here is that, assuming it’s the same skills keeping retention (which is admittedly a big assumption), this means people pick the skills they want in the first three months and then over the next two years cull down 4 of them that they didn’t end up finding useful. Using 7 skills in the past week is actually a much higher number than I think people would have landed on, assuming they weren’t counting the utility skills we talked about above (that Amazon provides not as skills, but directly baked in – like reminders, music, etc).
Ok, now onto the Synthetic update, which is admittedly shorter than the voice update…
What's Up In Synthetic Reality
> Read Genderless voice: A More Inclusive Future with Tech - I’ve shared articles on genderless voice before, and here’s an update: Q, a new genderless AI assistant is being developed by Copenhagen Pride and Virtue to address this, producing a scientifically gender-neutral voice as a potential future replacement for the acutely female and male voices in today’s systems. The tool was developed by ranking the voices of several people who identify as female, male, non-binary, and transgender. Thousands of study participants were then asked to rank these voices on a spectrum from male to female, allowing the researchers to home in on what they’ve identified as a gender-neutral voice.
> Check out Cross & Freckle – they did an experiment to see how much of a startup could be generated using AI. It started with T-shirts, whose designs are computer-generated. But then they extended that to the logo, the brand-name and the marketing copy. The shirts are real and you can really buy them.
That’s all for this week. Have a great weekend!
– Matt
Did someone forward you this email? Subscribe at
You can follow me on twitter @matthartman
Did you enjoy this issue?
In order to unsubscribe, click here.
If you were forwarded this newsletter and you like it, you can subscribe here.
Powered by Revue