Andrew Weisselberg from SoundHound AI

[00:00:20] Hello everyone and welcome to Kilowatt, a podcast about electric vehicles, renewable energy, autonomous driving, and technical issues. On today's episode we have an interview for you. Normally we would have a news story for you, but I am having nothing but trouble with my Apple Pro apps like Logic and Final Cut. Tons of problems. Don't know why, but I am. It's not great. Not super happy about it.

[00:00:49] The biggest issue is when I try to export something, it will absolutely just reboot my computer. Like it'll die and reboot my computer. And then in Logic, I've got some weird artifacting. Hopefully, I think I have it fixed now. But I don't know where it came from. I don't know why it's happening. It's just happening.

[00:01:13] So, oh, and then I had some of my interviews from, or all of my interviews for that matter, from CES on my AnchorWork M650 Lavalier mics. You can record directly to the mics. And I was moving those files over. And for some reason, and I don't know why, the mics lost connection. And when they regained connection, all of the files were gone.

[00:01:38] So, thankfully, I've got a lot of those files saved. And I had a backup. I had some redundancy built in. So we didn't lose everything. So that's good. But I'm going to have to Frankenstein some files together.

[00:01:58] Anyway, long story short, I've had a very frustrating Thursday, Friday, and today, Saturday, have been extraordinarily frustrating for me. Good news is, though, I do have an interview for you. And yes, it's one of those interviews that I had to Frankenstein together because I had half of the conversation saved. And then I had to take the other half from my backup recording that I did on my iPhone.

[00:02:25] So hopefully it sounds good. I think it sounds actually pretty good compared, you know, considering everything I had to do to get it to the sound that way. But yeah, yeah, it's just been an interesting day. So we are going to talk to Andrew of SoundHound AI. And if you don't know who SoundHound are, you should go to their website at SoundHound.com. But they are a voice assistant. They started off with a music app.

[00:02:51] And then they also had like a Siri like app back when Siri wasn't owned by Apple. So they have grown and we're going to hear they've gotten into automotive and they do drive throughs and kiosks and retail and all that stuff. So we're going to hear about what SoundHound is up to mostly in the automotive market. So let's go ahead and get to the interview. All right. I am here with Andrew of SoundHound AI. Andrew, how are you finding CES?

[00:03:20] I'm finding CES great. We got big crowds. We're showing off some new innovation, innovative voice AI technology. We just launched our voice commerce ecosystem where we're bringing together automotive and restaurant so that you can order food directly from your car without with only using your voice. And it's been a fantastic reception. And we're excited to be here.

[00:03:46] Yeah. And I was actually recorded like I got back from showstoppers at like, I don't know, midnight last night. And then I actually recorded an episode and I mentioned that you and SoundHound and Lucid are now integrated together. And I've said this to every single person I've talked to in this booth, but I think what SoundHound is doing is absolutely incredible. And I do not think that people realize how many places this company is, including, you know, their their vehicles. You want to talk a little bit about that?

[00:04:15] Yeah, no, absolutely. Well, SoundHound was founded in 2005. And the vision was for our three co-founders who all met at Stanford was looking out 20, 30 years. Like, what is an incredible innovation that's going to happen in our lifetime? They were big fans of Star Trek. So one of the things they looked at was voice AI technology, people talking to computers. That is going to happen in my lifetime. So they decided to start a company to pursue that objective.

[00:04:42] But the reality was the technology had to be built. It wasn't there yet. But so what they realized was, hey, we can launch an app, a music recognition app, because there's a real need there. People listen to music. Oh, what's that song? I don't know. You hold up an app, it recognizes it, tells you what it is, and then it pushes you over to the app store or a music store and you get it and buy that.

[00:05:03] So that was SoundHound Music. But what we were really doing was using that app, which has gotten over 300 million downloads over the last 20 years or so, using that app to build our voice AI technology stack. So what you see in automotive right now that you're asking about is really the full capability of what we've been building with our voice AI technology.

[00:05:28] So there's a few companies in the world that have a full stack voice AI platform. So that consists of automatic speech recognition, that consists of natural language understanding, and that consists of text to speech. Most people know the bigger companies that have these stacks. They're big tech, Amazon, Google, Apple. We are one of the very few companies that have that.

[00:05:54] And we've been able to successfully take that R&D where we were incubating it through that music app, because a lot of people, to your point, associate us with a music app. But what we were really doing was building a platform so that we could put voice AI technology in cars, products, devices. Now we're in restaurants.

[00:06:13] And on the enterprise side, we're starting to get into banking, insurance, healthcare, retail, basically every industry that is going to be affected by AI we are now playing in. That is fantastic. Fantastic. I find that more and more in my life, I do not have time to reply to a text or any, it's just something simple. But being able to talk to your car, and right now I have a Model Y, and I'll be honest with you, that's not a great experience to talk to my car and my Model Y.

[00:06:42] I've always, I've been saying on my podcast for years that Tesla should just give up the ghost and go with SoundHound or an equivalent company. But right now we're standing in front of a Lucid, looks like a Sapphire, Lucid Air Sapphire. What kind of integrations does SoundHound have with this? Because this was just recently announced, I think it was this week, right? Yeah, so it was announced this week.

[00:07:06] We actually featured a Lucid car in Detroit in October at Reuters Automotive USA. So keen eyes who were following us would have noticed that the Lucid announcement before because that was out there with the public. But yeah, I mean, our voice AI platform is called Houndify. So that is the technology. Again, that's the ASR, that's the NLU, that's the text-to-speech. So we have a platform that we go in with our engineers.

[00:07:36] We work closely with Lucid's engineers, and we basically customize the different use cases that they want for their car. So that can be opening the windows, opening the sunroof, changing the station on the radio. We call that the car control domains. But what's interesting about what we're showing at CES, and which is featured in this beautiful Lucid Air, is voice generative AI.

[00:08:00] So it's the integration of the large language models so that you can take the voice experience to the next level. So if you want to know what the real-time score of the basketball game is, you can ask that question. And not through Lucid Air, this is integrating ChatGPT, but perplexity, which we have other automotive customers that are leveraging that. You can do real-time and voice commerce.

[00:08:28] These are other things that are eventually going to happen, again, where you can order food. Basically, any question that you can ask a large language model really enhances the usability of these voice assistants. And I'll tell you, as the VP of marketing, we do a lot of research with end consumers. And what we found when we look at our data and we talk to customers is voice AI in the car has had very limited use cases.

[00:08:55] There's basically about 12 different things that people have done. And that kind of stagnated the business a little bit because auto manufacturers are like, okay, voice AI is great. I could connect to my phone. I could do navigation. These are use cases that people appreciated. But they really weren't keen to go beyond that because the consumers really weren't taking actions outside of those core use cases. With voice generative AI, which is featured in this Lucid, that's really breaking the mold.

[00:09:24] And we're seeing a great reaction over the last pretty much 12 months with auto manufacturers. Stellantis was the first one to launch it. We launched that with DS Auto in Europe. I believe that was in February. And that was the first commercial implementation of generative AI in an automobile in the world.

[00:09:47] And since that happened, we've just been getting great reactions from automakers because they're hungry for this technology because they're seeing it's expanding the use cases and they're using it as a competitive differentiator. Why is this car different than this car? Well, this car has amazing capabilities in voice AI. In our research, I could tell you that will make a difference between someone picking model A versus model B.

[00:10:12] If everything else is equal, but you have an amazing voice experience or you can order food in your car, for example, consumers are telling us, I will buy that car over the other one. How difficult of a job was it to convince automakers? Because honestly, voice commands versus talking to an LLM is a totally different thing.

[00:10:34] And when you're going in and maybe some of the folks that you're talking to are technical, I've actually found that most auto executives are extraordinarily technical in talking to them. Because they came from the line or they came from this engineering part of the company or whatever. So I don't want to paint it as if they're stupid people, but voice commands is not a very good way to do things versus an actual LLM, which could say, you know, turn the temperature up. And maybe you stumble on a word or something like that.

[00:11:03] The LLM could be like, I'm pretty sure you meant turn the temperature up versus I'm sorry, I can't do that today. I mean, when you go and you make that pitch, is that a difficult pitch to make or are they just kind of on board and they already kind of are familiar with what you're doing with LLMs and AI as a general rule? Well, when you talk to about auto manufacturers, you got to look at how they're incentivized. And generally speaking, they're incentivized in two ways. One is lowering costs.

[00:11:31] Second is increasing customer satisfaction. So clearly by implementing this, you know, very advanced AI technology in the car, it's not saving the automaker money. It's adding costs. So what does that mean? It means that you have to increase customer satisfaction. So what we found is very helpful where the auto manufacturers, they'll come in and we'll deal with customer experience people, product leaders, and they love the technology.

[00:11:57] But they have to sell it into their organization and say, why am I adding further costs onto this car? Sure. That's where we come in and really work with them. And we have this, you know, it's our internal code word, but we call it concept lab. But it's basically where we do deep research on not our customers, the automakers, but our customers' customers.

[00:12:21] So we can understand what are the use cases that are going to drive usage and drive customer satisfaction. So if we can bring that data back, whether it's initially through research, but then it's actually with some of the early adopters coming back and showing the data. For example, we saw usage increase 6x when we implemented voice generative AI in our first use cases with our first manufacturers.

[00:12:49] That 6x increase is off the charts. If I would have told an auto executive two years ago before they ever heard of generative AI that I have a solution that's going to increase usage by 6x, they wouldn't have believed, Jared. So that really is how we kind of work with the automakers to really kind of team up with them and help them kind of sell. Here's how we're going to drive that customer satisfaction. This is how you're going to help move more sheet metal.

[00:13:17] Because at the end of the day, they want to sell cars. We want to help them sell cars. But we want to do it with some cool technology that you're going to have the end user be delighted. And if you're focused on delightful customer experiences, we find in the end, usually automaker works out for us. And, you know, it's a win-win all around. It is. And I will say, one of the things that I do the most in my car when I'm by myself is I'll just turn ChatGPT on and I'll start talking to it.

[00:13:45] Because, like, there's generally four or five problems that I have in my life. Like, whether they're, like, technical, how to do this in a certain program, or I'm trying to overcome whatever it is. And I'll just noodle it out. And I don't know that ChatGPT, as a general rule, gives me, like, any definitive answers. But it's nice to just be able to talk. I would love, like, I can't afford a loose sitter, I'll be honest with you. But if they ever come out with the $50,000 model, then I'm in. But not at $79,000. But if I was able to sit in my car.

[00:14:14] They have very attractive lease prices, I will say. They do have very. All of the manufacturers have very attractive lease prices. Like, one of the things that I talk about most often on my show is, like, how all these companies are coming out with, like, 0% down and $200 a month for a lease. I'm like, how do you, I mean, they're just trying to get the cars out there. But anyway.

[00:14:37] Well, but I will bring up a point because, you know, where an automaker will get nervous, which we would get nervous, too, is if you just let the large language models kind of fly without any arbitration. That's where you run into an issue. So let me give you an example. Say you have car manufacturer A, and you have an open large language model as their voice assistant. And you say, what is the best car in the world?

[00:15:05] They're probably, a large language model is probably not going to say that the car you're in is the best. So an automaker is going to want to control that. So what we help automakers do is, through arbitration and allowing them to control the large language model, tame it, if you will, so that there is custom brand experiences that can be built, not by the engineers, but by the OEM's marketing team.

[00:15:29] And that is really what kind of pushes them through, because you don't want your customer to be scared of. If I put this technology in my product, whether it's a washing machine or an automobile or a mobile app, you don't want it to hallucinate and do damage to your brand. So that's one of the things that SoundHound AI is really good, because we have an amazing arbitration engine.

[00:15:55] It's really around how the software engineering comes in to arbitrate and say, okay, this is a query about turn on the radio and I want to listen to Lady Gaga, as opposed to let me tell you about Lady Gaga. It needs to arbitrate and decide which domain does that ultimately go to.

[00:16:14] So that's some of the secret sauce that really kind of differentiates better voice experience is, can you get the arbitration in a place where you can really minimize or completely eliminate those hallucinations that are going to create either a bad brand experience, in our case, for our customers, or it can create just a bad experience for the end user because they're not getting the answer that they're looking for.

[00:16:40] Right. It goes back to the problem again of voice control versus actual aspire. Yeah, you definitely have to have the guardrails in place because obviously GM doesn't want Ford saying, or doesn't want the car saying, oh, the Ford Mavic or whatever is the best pickup truck out there. Exactly. Which is a fantastic pickup truck that if you were GM, you don't want your voices sitting in your car talking about your competitors in that way.

[00:17:07] Or, you know, your car convincing you to, you know, hurt yourself or whatever the other nutty things these large language models do. But there is a nice sense of being able to have a conversation of like, oh, turn up the heat, follow that up with, oh, that's too hot. Give me two degrees less. Oh, and it's a safety issue too. You know, keep your eyes on the road. I mean, we all get distracted. You don't want to pick up your phone. You're fiddling with knobs. You're all these touchscreens.

[00:17:34] I mean, these touchscreens are fantastic. And, you know, you go into one of these new cars and it's like this, it's all you're surrounded by screen, but that is distracting. So it's not that you want to get rid of the screens in there. You need the voice control piece so that the safety ultimately is not sacrificed. And, you know, that's an important thing. I mean, I'm not going out there saying, hey, we're saving lives, but it's something we're really mindful of.

[00:17:59] And to be quite frank, going back to the original question of like, how did we get into automobiles? Automobiles were an early adapter of voice AI technology specifically for the safety reason. That's why we chose that industry first, because we thought they're going to be the early adopters because this is a safety issue. And, you know, I think that's an important thing. I agree with you. What other, because we talked about the Lucid, but what other car manufacturers are you working with?

[00:18:24] Yeah. So Hyundai, Stellantis, Lucid, Tog is a Turkish EV maker. There's a lot of other stuff that's going on that I can't talk about publicly, but overall we have about 20 different automotive brands and we're in millions of cars. That's amazing. Andrew, thank you so much. I really appreciate you coming on. Yeah, thank you. All right. I want to thank Andrew Weisselberg for coming on.

[00:18:53] Andrew, I should have mentioned this at the top. I thought I mentioned it in the actual interview, but I didn't. But Andrew is the VP of marketing for SoundHound AI. Again, if you want to see what they're up to, go to SoundHound.com. I'll put a link in the show notes as well. And yeah, Andrew, thank you for coming on and being so generous with this time, because there's a lot of people waiting to talk to these folks, just in case you were curious.

[00:19:19] When you go to CES and you do these interviews, there's like I had my interview cut short for Aptera because they already had a scheduled interview with Kyle from Outer Spec Studios. So yeah, there's just a ton of people trying to get at these people's time. So he spent about 16 minutes with me. Actually, it was more than that. It's more like 20 minutes, but the actual interview was like 16 minutes. So very generous. So Andrew, thank you so much for that. All right.

[00:19:49] I have before I let you go, I just have one little added thing to say. I on the last episode, I mentioned that Tesla, if you drive more than 50 percent of the time in full self-driving mode, you get 10 percent off your Tesla insurance. If you were one of those people and you lived in Arizona or Texas, that kind of thing.

[00:20:12] And I asked how many people are driving with full self-driving on, you know, the majority of the time, some of the time. What do you think your percentages are? And I'm quite surprised. I did not get anybody email in and say that they didn't use it very much. I did get quite a few people who said that they used it anywhere between 70 percent and 95 percent of the time.

[00:20:39] So that is that is I mean, it's not a high sample size, right? We didn't get a lot of responses, but still from the responses that we did get. Some of the people who responded, I wasn't really surprised. I was like, oh, yeah, I knew your usage was going to be high just based on previous conversations. But I didn't expect to have the lowest number be 70 percent of the time.

[00:21:03] By the way, we had a decent representation of hardware three and hardware four vehicles where most of the people told me what they were driving. So, yeah, interesting stuff. All right, everybody, that'll be it for this episode. As soon as I'm done and post this, I will be nuking and paving my laptop and starting all over again. So wish me luck because it is 645 at night. So I anticipate being at this until at least 11.

[00:21:33] But yeah, thanks, everybody, for being patient. I hope you enjoyed the interview and I will talk to you on Tuesday.