Alexa, Google, Scientist discussed on Inside VOICE
Of things. The second systems were would we in the industry typically call chat bots systems. They were very simple systems. That if you happen to say a keyword it would trigger response Those persistence came out in one thousand. Nine hundred sixty five and the chat pop. Boom that began about four years ago was based on these older technologies. The problem with these systems they can be very effective but most of them were not well built to understand what people were saying. A statistic that I saw yesterday was that there was only a twenty two percent user satisfaction rating of those voices systems. That use those primitive chat bots so a lot of people that have built companies in the past few years on. Those technologies are finding a lot of consumer pushback because they're just operating all that while and satisfying people and achieving the missions that they were set out to the third generation were systems like Siri and Alexa and Google Assistant. They began to improve the chat. Bot Technologies by adding machine learning. Now let's go up a level if you're in a scientist you have a lot of different tools available to machine. Learning is just one of them and for those of you who don't know much about machine learning it's using statistical analysis to take good gas at what someone is saying. It's a little bit like pattern matching if they hear a pattern of voice pattern of people saying something they can trigger a response the problem with these systems as good as a they are is. They're not very accurate. There was an interesting study that seating at did they release his past August. Doing an eight hundred question. Benchmark on all the primary systems things like Alexa scored seventy two and a half percent accuracy rating in this evening that test so it means that one out of four questions that you ask of. Alexa give a wrong answer and really I that syncs with my own experience. I use Alexa every day and when I request music very often it gets it wrong. So it's good for what it is and it's primarily good for non mission critical tasks like Music Selection. Or tell you what the weather is even the best system in the study Zd net found a google to only be eight percent accurate. That's pretty good but it's not good enough for most of our customers for instance when we're making a system for car and you want your window to go up and down your lights to turn on your seats go back and forth you need to have that. Work reliably every single time so We're building a system that is by the definitions in the presentation. I saw yesterday. A level for system level for system focuses not just on understanding commands but understanding people as they speak naturally conversation. Like you're talking to your best friend so that you can ask for your music in any way you might ask conversationally. An assist understands you. You can ask two questions at a time and a wander. Stand both of them separately and give you separate answers. It will understand the context of what you said and also it will be doing sentiment analysis to measure your motion as you're speaking because it turns out that emotion is also important in understanding what people are asking for or trying to achieve by working with voice assistant. So that's where we are today building level for system. Yeah and I love that you kind of broke that down because I have not thought about it that way before but it makes so much more sense and again really customizing it to you and I know your company has supplied conversational systems two cars motorcycles robots and a variety of consumer products. And you've built systems for companies like Bobo. Lg Mitsubishi and future robot. Can you talk about your process in developing conversational assistance for the automotive and transportation industry? Meaning what have you found works? Well and what are people liking most? What does the experience like so touching on something? I said a moment ago when we built our first assistant indycar about ten years ago there was a largely held belief in the auto industry. That voice interfaces didn't make any sense but that position has changed we have the Tesla factor going on in the auto industry and that is forcing all of the auto companies to think out of the box and to kind of reinvent how people use cars in interact with their cars their lot more receptive to new ideas. And there's two key things that we're trying to achieve in a car. I one is to make your diving experience more safe and I see a lot of interesting. Things will look at your existing car. You have to take your hand off the steering wheel to press a button to get something to work in your car or twist a map or push a lever. I would submit that anything that would cause you to take your hand off of the wheel and be distracted by that. Makes you less safe when you're driving so I do see new technologies like gestures in cars that make no sense at all to me? Why would you be even more distracted trying to make a gestures in the air with one of your driving hands once again? That's distracting you. Taking your hands off the wheel it makes great sense to me to use your voice to control your car. Yes I agree and I had seen a video that you did. I guess in the last year or so where you showcase kind of how it works. And you're asking voice assistant to move the seat up or to turn the D. fog or on and like you said it's something that can be done it's safe. It's easy it's effective. And it's customized to your vehicle in yourself. There's some important things that I think that we're doing. And maybe it's because my early start was doing three d interfaces and it was natural for us to put three talking. Characters with voice systems intuitively. We knew that people. If you made a good character they would bond with it and have a more satisfying experience with that voice system. If there was a character that went along with it but today were we've found there's been a lot of research done in the last ten years the confirms some of our hunches that is that people have higher user satisfaction higher trust scores higher engagement ratings and the National Highway Traffic Safety Administration even found that it increase safety by having a good visual interface that complements the voice interface. I think a lot of us have also heard an old statistic that seventy percent of all human communication is nonverbal. Why I don't exactly seventy percent but it strikes me that we in the voice industry are very focused on delivering a good experience for the thirty percent of human conversation. Let's face it. We're in the communication business. We want to effectively communicate with our users to understand what they want in order to better serve them and if we're only focused on the thirty percent solution were leaving. Seventy percent at risk so our view is if an Avatar can effectively reinforce the voice communication with appropriate facial gestures and interactions. That will work with what I call our caveman brain to have a complete communication and to more effectively communicate with the the driver or the user of the robot or person watching their TV interacting with a talking character on their TV. It's just hard wired into us to do this and what we're seeing in. The research is Improvements in the range of fifteen to twenty five percent in trust and user satisfaction scores by using good avatar. Yeah and that's interesting to note because I do see more of a push towards the multi modal where you're having the images with the voice but I know that there's also been a recent yell study where they found that voice. Only communication enhances empathetic accuracy relative to communication across various senses. Do you have any response to that particular study or do you still stand by. We're you saying voice and visual is equally as important together. I'm a person of science. I look at the research and in any topic of research. There are GonNa be contrarian views that come out if you look at the Avatar studies that were done ten to twenty years ago they all said the avatars are a bad idea. Well if you take a close look at the studies. They were doing cartoon. Stick figures as avatars. It's no wonder that their test subjects were reacting negatively to avatars. It's only really been in the past ten years where the study started using characters that were at the level of a good computer game or a a Pixar Disney movie that people started reacting positively to avatars and we took seriously ourselves The person on our team heading up avatars Alex Hessler. He came out to fix our he worked on the hesitate movie. It's important to us top talent whether it's the top. Ai Engineers or the top artistic talent to make the best possible things for our customers..