Lecture
Voice to Action: Building Portable Edge AI Applications with Linux and Containers
- at -
- Hall C5C5.100
- Language: English
- Type: Workshop
Lecture description
"What does it take to move an Edge AI application from a simple prototype to an architecture that can run across different embedded platforms?
In this interactive live session, we will build and evolve a voice-controlled Edge AI application together with the audience. Starting with a voice recognition model, an embedded Linux device will recognize spoken color commands and use them to control a physical light in real time.
The audience will then become part of the AI development process. Using only their smartphones and a shared link, attendees will contribute voice samples for a new color, creating a real dataset during the session. We will use Edge Impulse to train the updated model and deploy it as part of a containerized application. By the end of the session, the device will recognize a command that did not exist when the presentation started.
While the model is being trained, we will explore the architecture behind the demo: Yocto and Embedded Linux, containers, AI runtimes, hardware access, drivers and acceleration. We will discuss what can be portable inside a container, what remains dependent on the underlying platform, and why these boundaries matter for long-term product development and maintenance.
The same application will be shown on the Arduino® UNO Q, Arduino VENTUNO Q, Qualcomm Dragonwing™ IQ8 EVK, and Qualcomm Dragonwing IQ9 EVK platforms, illustrating how a common application architecture can scale from accessible development boards to professional Edge AI systems.
Rather than focusing only on inference, this session demonstrates the complete Edge AI development loop:
collect data → train → package → deploy → run on real hardware → improve
Attendees only need a smartphone to participate. All embedded hardware and live demonstrations are operated by the presenter.
What attendees will learn:
• How data for an embedded voice AI application can be collected collaboratively using smartphones
• How Edge Impulse simplifies dataset creation, training, and deployment for embedded AI
• How an AI model can be integrated into a containerized Embedded Linux application
• How Yocto, containers, AI runtimes, drivers, and hardware acceleration fit together
• What can remain portable across different embedded platforms and what stays hardware-specific
• How the same application architecture can scale from Arduino-class development boards to professional Edge AI platforms
• Why application/OS separation matters for updates and long-term product maintenance
No development board or laptop is required. Attendees will participate using their smartphones to contribute voice samples to the live dataset. All training, deployment, and embedded hardware interaction will be demonstrated live by the presenter."
In this interactive live session, we will build and evolve a voice-controlled Edge AI application together with the audience. Starting with a voice recognition model, an embedded Linux device will recognize spoken color commands and use them to control a physical light in real time.
The audience will then become part of the AI development process. Using only their smartphones and a shared link, attendees will contribute voice samples for a new color, creating a real dataset during the session. We will use Edge Impulse to train the updated model and deploy it as part of a containerized application. By the end of the session, the device will recognize a command that did not exist when the presentation started.
While the model is being trained, we will explore the architecture behind the demo: Yocto and Embedded Linux, containers, AI runtimes, hardware access, drivers and acceleration. We will discuss what can be portable inside a container, what remains dependent on the underlying platform, and why these boundaries matter for long-term product development and maintenance.
The same application will be shown on the Arduino® UNO Q, Arduino VENTUNO Q, Qualcomm Dragonwing™ IQ8 EVK, and Qualcomm Dragonwing IQ9 EVK platforms, illustrating how a common application architecture can scale from accessible development boards to professional Edge AI systems.
Rather than focusing only on inference, this session demonstrates the complete Edge AI development loop:
collect data → train → package → deploy → run on real hardware → improve
Attendees only need a smartphone to participate. All embedded hardware and live demonstrations are operated by the presenter.
What attendees will learn:
• How data for an embedded voice AI application can be collected collaboratively using smartphones
• How Edge Impulse simplifies dataset creation, training, and deployment for embedded AI
• How an AI model can be integrated into a containerized Embedded Linux application
• How Yocto, containers, AI runtimes, drivers, and hardware acceleration fit together
• What can remain portable across different embedded platforms and what stays hardware-specific
• How the same application architecture can scale from Arduino-class development boards to professional Edge AI platforms
• Why application/OS separation matters for updates and long-term product maintenance
No development board or laptop is required. Attendees will participate using their smartphones to contribute voice samples to the live dataset. All training, deployment, and embedded hardware interaction will be demonstrated live by the presenter."
