How to use Deep Learning to process audio data and classify sounds.
summary
Converts raw WAV into log‑mel spectrograms with librosa, applies augmentations, then trains a CNN from scratch and fine‑tunes pretrained vision/audio models (TensorFlow/Facebook/Transformers stacks). Uses common datasets (10 classes—short clips from urban/speech corpora like Speech Commands/City sounds), trains for ~3–4 epochs, and reports close training/validation losses and high accuracy with confusion‑matrix evaluation. Shows code-level steps (loading, FFT→spectrogram, feature maps) and that transfer learning saves time and improves accuracy.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:04Agenda and goals
- 2:39Converting audio to digital
- 4:19Spectrograms and CNNs
- 6:06Sound classification basics
- 7:11Dataset preparation
- 8:09Librosa audio features
- 11:24Logarithmic scale for clarity
- 12:15Building CNN from scratch
- 13:44Model performance assessment
- 14:48Transfer learning benefits
- 16:10Pretrained models overview
- 17:23Training setup and metrics
- 18:32Conclusion and libraries
- 24:45Q&A session
related talks
25mLooking at Sound
40mGathering Insights from Audio Data
22mImplementation and analysis of deep learning models for speech classification
01h 14mExploring the hidden potential of sound data
22mDeep learning module recognizes video sounds
45mThe Sound of Your Secrets: Teaching Your Model to Spy So You Can Learn to Defend
44mNeural Networks in Sound Analysis and Generation
20mExploring the hidden potential of sound data