After Censorship, Self-Censorship… but This Time It’s Funny, ...
summary
A prototype bleep censor converts audio to spectrograms and detects target sounds (applause vs ambient) by applying transfer learning: feeding spectrogram images into a pre‑trained CNN and training only the final layer. Training is implemented with Keras/TensorFlow (demoed via TensorFlow.js), and live inference runs in the browser using TensorFlow.js with the WebGL backend; training is fast for a single class and inference is lightweight on mobile, though tuning is needed to reduce false positives.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 2:07Motivation
- 9:54Prototype tools
- 13:45Neural networks explained
- 21:56Interactive demo
- 28:29Q&A on training
- 31:17Summary
related talks
01h 14mExploring the hidden potential of sound data
45mThe Sound of Your Secrets: Teaching Your Model to Spy So You Can Learn to Defend
20mExploring the hidden potential of sound data
25mLooking at Sound
28mHow to use Deep Learning to process audio data and classify sounds.
39mMachine Learning Magic for Your JavaScript Application
26mGettin’ into the (Tensor)Flow
34mA Powerful Introduction to Machine Learning with Js