vhug/vishal_voice_cloning_15secs · Datasets at Hugging Face

audio audioduration (s) 10.3 17.5	text stringlengths 139 302
	Erhard Krohbisky Reviewer. So, artificial intelligence, self-driving user experience. How is artificial intelligence leveraged in the automotive industry?
	I would like to start my talk with this interesting quote from Matt Ridley from his recent book, How Innovation Works, which is about innovation itself. Innovation happens when people are free to think, experiment, and speculate.
	The modern car needs to be rethought. Connectivity. It needs to be a part of our digital life. Software-driven. It needs to be an intelligent companion.
	Electrification. The mode of transportation needs to be sustainable. And autonomous driving. The passenger needs to be driven autonomously, but it also needs to be safe and comfortable. Labreet. Good morning.
	My name is Vishal. I'm 26 years old, and I work as an applied AI engineer at Karyad SE. We at Karyad are about 4,500 colleagues working towards the transformation of automotive mobility.
	And at Karyad, I work at the intelligent cockpit and body domain, and one of our main goals here is to develop holistic in-car experiences that power the Volkswagen group vehicles.
	I'm here to talk about three different things and how they work with each other. First, about artificial intelligence, a gentle introduction, followed by its state and the industry. Then, autonomous driving, how artificial intelligence is used for autonomous driving.
	And finally, about user experience, how the car is perceived as a digital product. And also, I'm here to show some use cases and concepts which are powered by AI.
	Moving on, the Oxford Dictionary defines artificial intelligence as the theory and development of computer systems that perform specific tasks which require human-level intelligence.
	And these tasks range from visual perception to speech recognition, decision-making, and also translating languages. AI is further classified or categorized into artificial general intelligence,
	where the intelligence is on par with humans. The software component has a self-aware consciousness. It can learn, adapt, and plan for the future. Some of the fictitious examples that we have come across in the past include
	HAL from Space Odyssey and Jarvis from Iron Man. But are we there yet? DALI is an artificial general intelligence software which was produced or released by OpenAI,
	a research laboratory which works dedicatedly towards artificial general intelligence applications. What this AGI does is create photorealistic images from textual descriptions.
	So when the AI was asked to create an avocado-shaped chair, this was the result. And it looks photorealistic. It's pretty close and real.
	And as we just talked, another artificial general intelligence concept was released in the last days by DeepMind, which goes by the name Gato, which is a translation for cat in Spanish.
	And now this artificial general intelligence can do many more tasks all at the same time, or one after the other, with just one model. And these tasks range from controlling a robot to answering questions and also analyzing pictures.
	And the other category is artificial narrow intelligence, which is trained on a specific task. It is domain-specific and cannot generalize. So an AI which was developed for one specific purpose cannot be recycled for another purpose right out of the box.
	And some of the day-to-day use cases that we come across in our lives include Google Assistant, Alexa, Siri. And in an automotive context, if we want to detect if the driver is drowsy, drowsiness recognition. We can use an AI to do that.
	What is behind an AI and how does it work? So the term artificial intelligence is interchangeably used with the word machine learning, which is essentially a subset of AI for performing specific tasks. And machine learning is essentially pattern matching.
	You learn matching patterns from large amounts of data, detect new patterns and associate them with previous knowledge that you already know, and finally you make predictions. So when such an image of a road is subjected to an AI,
	which has seen many such roads in the past during the training phase, one of the first steps is to break it down into simpler forms, and this is done through convolutional neural networks, and that's what you see there. And finally, the underlying pattern is understood and a prediction is made.
	In this case, the AI now knows it's a road. The McKinsey Global Survey on AI served as a really good indicator to understand the state of AI in the industry.
	So this survey saw about 1,843 organizations representing different industries coming from emerging and developed countries. So one of the most important things that the survey reported
	was the adoption of AI in the year 2021, despite of the pandemic. 57% of the participants who took part in the survey acknowledged the adoption of AI in at least one of their functions,
	and some of the top use cases being service operations optimization to AI-based product feature enhancement and contact center automation. And looking at the success factors
	behind some of the best AI outperformers in the market, it came to the notice that these AI performers follow best practices for AI development, and this led them to have at least 20% bottom-line growth.
	And they claimed the following two be the best practices, to develop AI on the cloud and have data quality protocols. As I mentioned, AI sees lots of data in its training phase, and these companies claim that they have quality protocols
	to check that the data makes sense. And when these organizations had also asked about the risk that they face in bringing the AI products to use in their functions, they claimed the following four to be the biggest risks.
	Cybersecurity risk, regulatory compliance risk, explainability risk, and model documentation. I want to explain why explainability risk is really important for an automotive use case, because when you delegate the most important task
	of driving to a computer, you need to know the reason behind its decision-making. So in the event of an accident, the situation can be judged better. Artificial intelligence is used in the automotive industry
	for value creation in really different domains. In connectivity, AI is used for cybersecurity, intrusion detection. In in-car functions, AI is used for speech recognition, user profile management.
	In automated driving, AI is used for image processing in different stages, like vehicle perception and localization, some of which we will see in the next slides. In vehicle energy, AI is used for drivetrain energy management
	and intelligent road planning. Finally, in the mobility services, AI is used for predictive vehicular maintenance. Machine learning algorithms power a self-driving car's technology stack,
	right from cybersecurity detection to object detection to sensor calibration to also finally providing the user with the best user experience. And essentially, these machine learning algorithms fall into three different stages. Perception,
	localization, and motion planning. Understanding the scene is like one of the most fundamental tasks of an autonomous vehicle. And it does this through its sensor suit of cameras, lidars, radars,
	inertial measurement units, global position systems, and so on. In the perception stage, this is responsible for perceiving and identifying what is around a car. It is through this stage that a car knows all the different things in its environment,
	be it other cars in the environment or traffic signs or drivable areas and pedestrians. On the right, you see how objects are getting detected in a 2D image space and also in 3D point cloud data from lidars.
	The localization stage is responsible for providing the vehicle with its accurate location because the autonomous vehicle has to know its location throughout its full journey. The perception data which we just saw in the previous stage is fused with geospatial information
	which you get from your GPS systems. And finally, you prepare high-definition maps and these maps have centimeter-level accuracy. The motion planning stage is responsible for planning the vehicle's latitudinal and longitudinal motion.
	It is this stage which makes all the decisions depending on the dynamics of the environment. Let's say there's an ambulance driving past. Now the autonomous vehicle has to know what trajectory it has to follow. Or if a car has to be overtaken, again, a trajectory has to be determined
	that the autonomous car has to follow. Finally, when you combine all the different information from the three stages, you have a multimodal object detection and the process is called sensor fusion which does this and this is how it looks
	like what you see on the right. All the different information from the scene is fused together to give the car a better understanding of the scene. I want you all to look at the four graphic elements on the right side. What do you see?
	I see triggered emotions. Engagement, excitement, and curiosity. But most importantly, these emotions occur to the user even before starting to use the product. Hence,
	the intrinsic expectations from a digital product has dramatically increased in the last few years. But why is user experience and why is intrinsic expectation relevant for a self-driving car? You might ask.
	The modern car, as we all know it, is an intelligent companion. It needs to be connected. it is autonomously driven and also is electrified. The ability to delegate the driving task to a computer now makes everyone
	inside a car a passenger. So satisfying the intrinsic expectations that I just mentioned in the previous slide, like emotions, is more relevant now than it was before. And artificial intelligence is the bridging gap that helps us in satisfying
	these intrinsic expectations. So the car now becomes a feature-rich digital product. What you see here on the left is how the light sequence is welcoming its user in a car. And on the right side,
	you see an Audi A6 concept where the headlamp projectors are used for gaming when you're charging your car at a station. No wonder the consumer electronic shows of the past have been stormed by automakers. Personalization.
	There are two sides to this coin. First, personalization makes the user welcomed. second, the personalization makes the product unique. The priority boating feature recognizes the gate of an owner approaching the vehicle
	and it automatically opens the door so he has a seamless onboarding. And since now we know that the occupant that's entering the car, we can make personalized in-car adjustments for this occupant. What you see here on the right is how the car is recognizing
	someone walking towards it and tries to open the door for him. It also knows the person who is walking towards it. What are the benefits you may ask? Automatic personalization of the cabin. So now we know who the user is,
	we can have a personalized welcome for him and we can also alter his seating positions based on his previous user configurations and also the temperature can be changed. Many more such features can be added which I might also not know at this point but the possibilities
	are limitless. Finally, since we know who the user is, we can also integrate well with his digital ecosystem. For example, route suggestions can be made based on his daily activities, third-party integration to calendar music
	and so on. Productivity. This is really important for all of us because this is what is progressing each of us towards our full potential and being productive is a badge that we all want to wear.
	The video conferencing feature provides the user with a hands-free best-in-class video conferencing so he can use his time during commute for better reasons. What are the benefits you may ask? Productivity, as I just mentioned,
	you can have a hands-free video conferencing and use your time effectively when you're traveling. Again, personalization plays another role here because depending on the seating of the occupant, you can have
	an occupant-zone-based audio management which will let different users in a car have a video conference at the same time independently. And we all are used to scene enhancements in the past years
	because of the pandemic where we got used to different video conferencing tools like Microsoft Teams, Zoom, and so on. You can also have virtual backgrounds and synthetic avatars. The last concept that I want to show is a really interesting one which is about
	gamification. HoloRide provides the passengers with a hyper-immersive experience using virtual reality. So what you see here on the right is how a virtual reality goggle is completely synced with the vehicle's motion
	and it's not just the vehicle's motion but it's also the environment. The pedestrians that you saw were recognized as characters in the game that the user is playing. What does this provide? Engagement.
	It provides the user with the most immersive digital experience using elastic adaptable content. And entertainment. You can entertain the user by providing him with on-demand content which he can purchase
	based on his route. Let's say you want to entertain your kid with a planetarium package. You just download it and he puts his virtual goggle on and yeah, he has an immersive planetarium experience. And it's realistic
	because everything is motion synced and it's traffic and location aware. That brings us to the end of my talk. I hope this was interesting. I'm open for questions. You can reach out to me later. Thank you. Thank you. Thank you, Vishal. Wow.
	Wonderful presentation. Can you turn my Audi into a TIE fighter so I can put on VR glasses and enjoy my Star Wars while driving? Yeah. That's the future. Excellent. Thank you very much. We'll speak to you in the discussion later on. Thank you. Thank you, Vishal. I'll turn those

Erhard Krohbisky Reviewer. So, artificial intelligence, self-driving user experience. How is artificial intelligence leveraged in the automotive industry?

I would like to start my talk with this interesting quote from Matt Ridley from his recent book, How Innovation Works, which is about innovation itself. Innovation happens when people are free to think, experiment, and speculate.

The modern car needs to be rethought. Connectivity. It needs to be a part of our digital life. Software-driven. It needs to be an intelligent companion.

Electrification. The mode of transportation needs to be sustainable. And autonomous driving. The passenger needs to be driven autonomously, but it also needs to be safe and comfortable. Labreet. Good morning.

My name is Vishal. I'm 26 years old, and I work as an applied AI engineer at Karyad SE. We at Karyad are about 4,500 colleagues working towards the transformation of automotive mobility.

And at Karyad, I work at the intelligent cockpit and body domain, and one of our main goals here is to develop holistic in-car experiences that power the Volkswagen group vehicles.

I'm here to talk about three different things and how they work with each other. First, about artificial intelligence, a gentle introduction, followed by its state and the industry. Then, autonomous driving, how artificial intelligence is used for autonomous driving.

And finally, about user experience, how the car is perceived as a digital product. And also, I'm here to show some use cases and concepts which are powered by AI.

Moving on, the Oxford Dictionary defines artificial intelligence as the theory and development of computer systems that perform specific tasks which require human-level intelligence.

And these tasks range from visual perception to speech recognition, decision-making, and also translating languages. AI is further classified or categorized into artificial general intelligence,

where the intelligence is on par with humans. The software component has a self-aware consciousness. It can learn, adapt, and plan for the future. Some of the fictitious examples that we have come across in the past include

HAL from Space Odyssey and Jarvis from Iron Man. But are we there yet? DALI is an artificial general intelligence software which was produced or released by OpenAI,

a research laboratory which works dedicatedly towards artificial general intelligence applications. What this AGI does is create photorealistic images from textual descriptions.

So when the AI was asked to create an avocado-shaped chair, this was the result. And it looks photorealistic. It's pretty close and real.

And as we just talked, another artificial general intelligence concept was released in the last days by DeepMind, which goes by the name Gato, which is a translation for cat in Spanish.

And now this artificial general intelligence can do many more tasks all at the same time, or one after the other, with just one model. And these tasks range from controlling a robot to answering questions and also analyzing pictures.

And the other category is artificial narrow intelligence, which is trained on a specific task. It is domain-specific and cannot generalize. So an AI which was developed for one specific purpose cannot be recycled for another purpose right out of the box.

And some of the day-to-day use cases that we come across in our lives include Google Assistant, Alexa, Siri. And in an automotive context, if we want to detect if the driver is drowsy, drowsiness recognition. We can use an AI to do that.

What is behind an AI and how does it work? So the term artificial intelligence is interchangeably used with the word machine learning, which is essentially a subset of AI for performing specific tasks. And machine learning is essentially pattern matching.

You learn matching patterns from large amounts of data, detect new patterns and associate them with previous knowledge that you already know, and finally you make predictions. So when such an image of a road is subjected to an AI,

which has seen many such roads in the past during the training phase, one of the first steps is to break it down into simpler forms, and this is done through convolutional neural networks, and that's what you see there. And finally, the underlying pattern is understood and a prediction is made.

In this case, the AI now knows it's a road. The McKinsey Global Survey on AI served as a really good indicator to understand the state of AI in the industry.

So this survey saw about 1,843 organizations representing different industries coming from emerging and developed countries. So one of the most important things that the survey reported

was the adoption of AI in the year 2021, despite of the pandemic. 57% of the participants who took part in the survey acknowledged the adoption of AI in at least one of their functions,

and some of the top use cases being service operations optimization to AI-based product feature enhancement and contact center automation. And looking at the success factors

behind some of the best AI outperformers in the market, it came to the notice that these AI performers follow best practices for AI development, and this led them to have at least 20% bottom-line growth.

And they claimed the following two be the best practices, to develop AI on the cloud and have data quality protocols. As I mentioned, AI sees lots of data in its training phase, and these companies claim that they have quality protocols

to check that the data makes sense. And when these organizations had also asked about the risk that they face in bringing the AI products to use in their functions, they claimed the following four to be the biggest risks.

Cybersecurity risk, regulatory compliance risk, explainability risk, and model documentation. I want to explain why explainability risk is really important for an automotive use case, because when you delegate the most important task

of driving to a computer, you need to know the reason behind its decision-making. So in the event of an accident, the situation can be judged better. Artificial intelligence is used in the automotive industry

for value creation in really different domains. In connectivity, AI is used for cybersecurity, intrusion detection. In in-car functions, AI is used for speech recognition, user profile management.

In automated driving, AI is used for image processing in different stages, like vehicle perception and localization, some of which we will see in the next slides. In vehicle energy, AI is used for drivetrain energy management

and intelligent road planning. Finally, in the mobility services, AI is used for predictive vehicular maintenance. Machine learning algorithms power a self-driving car's technology stack,

right from cybersecurity detection to object detection to sensor calibration to also finally providing the user with the best user experience. And essentially, these machine learning algorithms fall into three different stages. Perception,

localization, and motion planning. Understanding the scene is like one of the most fundamental tasks of an autonomous vehicle. And it does this through its sensor suit of cameras, lidars, radars,

inertial measurement units, global position systems, and so on. In the perception stage, this is responsible for perceiving and identifying what is around a car. It is through this stage that a car knows all the different things in its environment,

be it other cars in the environment or traffic signs or drivable areas and pedestrians. On the right, you see how objects are getting detected in a 2D image space and also in 3D point cloud data from lidars.

The localization stage is responsible for providing the vehicle with its accurate location because the autonomous vehicle has to know its location throughout its full journey. The perception data which we just saw in the previous stage is fused with geospatial information

which you get from your GPS systems. And finally, you prepare high-definition maps and these maps have centimeter-level accuracy. The motion planning stage is responsible for planning the vehicle's latitudinal and longitudinal motion.

It is this stage which makes all the decisions depending on the dynamics of the environment. Let's say there's an ambulance driving past. Now the autonomous vehicle has to know what trajectory it has to follow. Or if a car has to be overtaken, again, a trajectory has to be determined

that the autonomous car has to follow. Finally, when you combine all the different information from the three stages, you have a multimodal object detection and the process is called sensor fusion which does this and this is how it looks

like what you see on the right. All the different information from the scene is fused together to give the car a better understanding of the scene. I want you all to look at the four graphic elements on the right side. What do you see?

I see triggered emotions. Engagement, excitement, and curiosity. But most importantly, these emotions occur to the user even before starting to use the product. Hence,

the intrinsic expectations from a digital product has dramatically increased in the last few years. But why is user experience and why is intrinsic expectation relevant for a self-driving car? You might ask.

The modern car, as we all know it, is an intelligent companion. It needs to be connected. it is autonomously driven and also is electrified. The ability to delegate the driving task to a computer now makes everyone

inside a car a passenger. So satisfying the intrinsic expectations that I just mentioned in the previous slide, like emotions, is more relevant now than it was before. And artificial intelligence is the bridging gap that helps us in satisfying

these intrinsic expectations. So the car now becomes a feature-rich digital product. What you see here on the left is how the light sequence is welcoming its user in a car. And on the right side,

you see an Audi A6 concept where the headlamp projectors are used for gaming when you're charging your car at a station. No wonder the consumer electronic shows of the past have been stormed by automakers. Personalization.

There are two sides to this coin. First, personalization makes the user welcomed. second, the personalization makes the product unique. The priority boating feature recognizes the gate of an owner approaching the vehicle

and it automatically opens the door so he has a seamless onboarding. And since now we know that the occupant that's entering the car, we can make personalized in-car adjustments for this occupant. What you see here on the right is how the car is recognizing

someone walking towards it and tries to open the door for him. It also knows the person who is walking towards it. What are the benefits you may ask? Automatic personalization of the cabin. So now we know who the user is,

we can have a personalized welcome for him and we can also alter his seating positions based on his previous user configurations and also the temperature can be changed. Many more such features can be added which I might also not know at this point but the possibilities

are limitless. Finally, since we know who the user is, we can also integrate well with his digital ecosystem. For example, route suggestions can be made based on his daily activities, third-party integration to calendar music

and so on. Productivity. This is really important for all of us because this is what is progressing each of us towards our full potential and being productive is a badge that we all want to wear.

The video conferencing feature provides the user with a hands-free best-in-class video conferencing so he can use his time during commute for better reasons. What are the benefits you may ask? Productivity, as I just mentioned,

you can have a hands-free video conferencing and use your time effectively when you're traveling. Again, personalization plays another role here because depending on the seating of the occupant, you can have

an occupant-zone-based audio management which will let different users in a car have a video conference at the same time independently. And we all are used to scene enhancements in the past years

because of the pandemic where we got used to different video conferencing tools like Microsoft Teams, Zoom, and so on. You can also have virtual backgrounds and synthetic avatars. The last concept that I want to show is a really interesting one which is about

gamification. HoloRide provides the passengers with a hyper-immersive experience using virtual reality. So what you see here on the right is how a virtual reality goggle is completely synced with the vehicle's motion

and it's not just the vehicle's motion but it's also the environment. The pedestrians that you saw were recognized as characters in the game that the user is playing. What does this provide? Engagement.

It provides the user with the most immersive digital experience using elastic adaptable content. And entertainment. You can entertain the user by providing him with on-demand content which he can purchase

based on his route. Let's say you want to entertain your kid with a planetarium package. You just download it and he puts his virtual goggle on and yeah, he has an immersive planetarium experience. And it's realistic

because everything is motion synced and it's traffic and location aware. That brings us to the end of my talk. I hope this was interesting. I'm open for questions. You can reach out to me later. Thank you. Thank you. Thank you, Vishal. Wow.

Wonderful presentation. Can you turn my Audi into a TIE fighter so I can put on VR glasses and enjoy my Star Wars while driving? Yeah. That's the future. Excellent. Thank you very much. We'll speak to you in the discussion later on. Thank you. Thank you, Vishal. I'll turn those