Below is the current agenda for the November 16, 2018 event
(subject to updates and changes)
Doors Open. Welcome to the 7th Kranky Geek event! The event will start promptly at 10:00 am.
Welcome and event overview from Kranky Geek!
Discord will share the engineering and product decisions that enabled us to massively scale our operations with a small team and limited resources. We will focus on WebRTC modifications that enabled us to improve user experience for gamers.
Facebook’s Portal is a WebRTC-based video calling gadget where users have a smart self guided camera, which can also be controlled by the receiver, but ultimately still needs to interop nicely with other devices in order to leverage the larger reach, without sacrificing quality. Here’s how Facebook met that challenge.
As video streams grow in complexity and users seek deeper analysis of their contents, new cloud tools are needed to handle increasing processing needs. Intel will discuss some newly-open-sourced visual processing tools, including codecs, for the accelerated growth of the visual cloud.
Our speech contains so much more than the words we speak – tone, pitch, cadence, volume and other factors give signals that sometimes carry just as much information if not more than the speech itself. Walter will discuss some of these metrics, how to capture them, and how they can be used in speech analytics applications.
Understanding and transcribing speech in a real-world, noisy meeting environment is no simple task. Hear Ahmad speak to the multiple ways he has explored the ability to train machines to understand the words spoken and the multiple layer engines applied to improve accuracy.
Breakfast, lunch, snacks, and coffee/tea/water will be served during the event. Vegetarian, gluten free, and vegan meals will be served. You are also welcome to bring your own food, if you’d like.
Kranky Geek is proud to sponsor Girl Develop It! Lets give them a moment to explain their mission.
This talk will take a look under the hood at some of the ways AI is being injected into call centers including voice BOTs that understand natural language and cognitive assistants that help reduce call hold times with live agents. Brian will also raise some of the challenges that are commonly encountered along the way.
Every industry and organization has its own unique vocabulary. Speech analytics engines trained on a standard dictionary will never get these words right even though they are often some of the most important ones to catch. Etienne will discuss how to design and build a custom language model without intensive retraining for every environment.
Consumers are used to talking to voice assistants like Alexa and Siri, yet they get awful touchtone interfaces when they call customer service. That is about to change as voicebots powered by popular bot Natural Language Understanding (NLU) engines meet RTC. Nexmo will give a walkthrough on how to build a conversational IVR in minutes using Dialogflow as an example.
WebRTC produces mountains of data that can be used to optimize streams and debug problems – if you know where to look. This talk describes how AI and ML techniques are used to analyze RTC metrics and overcome key challenges, including identifying the root cause of communication impairments, determining where to put your infrastructure, and adjusting end-point configurations in real time for changing network conditions
There are many established methods for measuring media quality, but accuracy of these systems often varies based on the end point manufacturer and model. RingCentral will demonstrate how it uses Machine Learning (ML) with Deep Neural Networks (DNN) to predict stream quality in highly variable environments with 99% accuracy. The session will examine the methodology and how it applies to telephony networks and beyond.
Google will give an update on its contributions to WebRTC and the status of WebRTC 1.0 in Chrome. The presentation will feature a deep-dive into the soon-to-be-completed architectural changes in Chrome aimed to make real time audio more robust and echo free.
Get some food, ask some questions, make some connections.
Deep learning algorithms have been employed in many areas of real time video communications and broadcast. Small platforms such as mobile phones and tablets present significant challenges for the deep learning algorithms to achieve the desired performance and complexity. This study presents ways to alleviate the challenges by analyzing and optimizing the learnability of a deep neural network, and results are demonstrated.
ML has been successfully applied for audio and video applications but the usage in real time applications adds some new challenges. In the last years mobile OS providers have started to provide ML capabilities(CoreML, MLKit…) to overcome some of those limitations with on-device ML processing. Gustavo Garcia from Houseparty will review the status and capabilities of some of those APIs with a few examples of how to leverage them.
Mixed and Augmented reality aren’t just for games – XR is being used today for assisting remote field workers. In this talk James will discuss the challenges of streaming MR/AR with WebRTC using the HoloLens as an example
After Hours reception, with great wines from Grant Marie Winery, cocktails (including non alcoholic) from Alcademics, and live DJ music by Smoke Jumper
Many thanks to these sponsors for their economic & intellectual contribution to Kranky Geek worldwide events.
Beverages for Kranky Geek supplied by Grant Marie Winery and the mixologists at alcademics
DJ music by Smoke Jumper
Meet the great speakers at this year’s event
The Kranky Geek Event is organized by