tinyml2021: Presentation skeleton

jonnor · Feb 24, 2021 · bbadd73 · bbadd73
1 parent adac971
commit bbadd73
Show file tree

Hide file tree

Showing 7 changed files with 824 additions and 16 deletions.
diff --git a/tinyml2021/Makefile b/tinyml2021/Makefile
@@ -0,0 +1,8 @@
+
+slides.html: Makefile presentation.md style.css
+	pandoc -t revealjs -s presentation.md -o slides.html --slide-level=2 --mathml -V theme=white
+
+slides.pdf: Makefile presentation.md
+	pandoc -t beamer -s presentation.md -o slides.pdf --slide-level=2 --mathml --dpi 144
+
+all: slides.html slides.pdf
diff --git a/tinyml2021/README.md b/tinyml2021/README.md
@@ -1,5 +1,5 @@
 
-# TinyML2020
+# TinyML Summit 2021
 
 https://tinyml.org/home/index.html
 

diff --git a/tinyml2021/coffeecracking.md b/tinyml2021/coffeecracking.md
@@ -1,28 +1,19 @@
-# Acoustic Event Detection of coffeebean cracking during roasting
+# Acoustic detection of coffeebean cracking during roasting
 
 ## Talk proposal
 
 The key to a great coffee is high quality coffee beans
 and a roasting process that brings out the desired flavor and aroma.
 
-In this talk you will hear about how Soundsensing and Roest have developed an
-automated system that ensures consistent roasting of coffee.
-The system uses a MEMS microphone to pick up the sound of the roasting process,
-and a Machine Learning model to detect in real-time the sound of coffee beans popping ("cracking").
-The model runs on a custom developed board using an STM32 microcontroller
-and the X-CUBE-AI machine learning toolkit.
+In this talk you will hear about how Soundsensing and Roest have developed an automated system that ensures consistent roasting of coffee. The system uses a MEMS microphone to pick up the sound of the roasting process, and a Machine Learning model to detect in real-time the sound of coffee beans popping ("cracking"). The model runs on a custom developed board using an STM32 microcontroller and the X-CUBE-AI machine learning toolkit.
 Roest coffee machines have been shipping with this TinyML-powered feature since August 2020.
 
-We will cover the development process from the start and all the way to a validated solution,
-including system design, data collection, model development, and testing.
+We will cover the development process from the start and all the way to a validated solution, including system design, data collection, model development, and testing.
 
 Roest is a leading manufacturer of coffee-roasting machines.
-Their sample roaster is used by high-end coffee makers all over the world,
-and is the first commercially available coffee roaster that uses acoustic sensing.
+Their sample roaster is used by high-end coffee makers all over the world, and is the first commercially available coffee roaster that uses acoustic sensing.
 
-Soundsensing is a leader in acoustic monitoring using Machine Learning.
-Their IoT sensors are used to monitor noise levels in workplaces and cities,
-and to monitor the condition of machines and processes in manufacturing and real-estate.
+Soundsensing is a leader in acoustic monitoring using Machine Learning. Their IoT sensors are used to monitor noise levels in workplaces and cities, and to monitor the condition of machines and processes in manufacturing and real-estate.
 
 ## Call-to-Action
 Have an sensing/monitoring problem that can be approached with sound?

diff --git a/tinyml2021/environmentalsoundclassification.md b/tinyml2021/environmentalsoundclassification.md
@@ -27,7 +27,6 @@ Be it in cities, residential areas, offices, or factory floors.
 Order the Soundsensing dB20 devkit,
 and join our pilot program for Noise Classification.
 
-
 
 ## Has talk been given before
 Kind-of yes, kind of no. EuroPython 2019. But very little focus on optimization/microcontroller

diff --git a/tinyml2021/notes.md b/tinyml2021/notes.md
@@ -0,0 +1,81 @@
+
+12 minutes speaking time.
+TED style
+
+1. Main point
+TinyML is challenging but worth it
+
+Around 12-24 slides.
+
+
+# Done
+- Registered with Jotform
+
+# PDF export
+
+With reveal.js, can add  ?print-pdf to URL
+Note: must go before the #part
+
+# TODO
+
+- Import slides from ?
+- Make slide size fit TinyML. Page size 13.34 × 7.5 inches
+Default with Beamer was 5.04 × 3.78 inch
+Default with Chrome save to PDF was 20.00 × 11.25 inch
+1.777 aspect ratio
+When using Chrome print, system dialog. Can use custom paper format.
+Worked to get 13.34 × 7.5 inches
+
+- Add TinyML logo to title slide. Which logo? Email sent
+
+- Make slides respect outline 
+- Send to organizers.
+Bette & Rosina
+Release form. DONE
+Mobile phone number. DONE
+Slides outline
+
+
+# Outline
+
+### Intro
+Me. Jon Nordby
+Soundsensing company
+
+### Noise
+The problem
+
+### Noise Monitoring
+IoT sensors.
+Why TinyML
+
+### Audio Classification
+Pipeline
+
+### Environmental Sound Classification
+Urbansound8k
+
+## CNNs for efficient audio classification
+Slides from SenseCamp
+
+1,2,3,4 slides
+
+## Research results
+From thesis
+
+Feasible areas
+
+## In real world
+PNB. Activity detection. Noise at population
+
+Trafficmix. Noise separated by source
+
+
+# Misc
+
+Outputs
+Audio Classification as first piece of bigger systems
+Bringing to market
+Deployments
+Challenges
+