Lecture 9 - Sound processing & speech recognition

Teacher: Julie Mauclair (IRIT)

Contents

• How to encode audio ?

  • Raw signal / sampling / DFT /iDFT.
  • Mel Frequency Cepstral Coefficients (MFCC).

• Automatic speech recognition (ASR) from 1970 to 2010.

  • Isolated word recognition.
  • Noisy channel models.
  • Standard ASR systems.

• Automatic speech recognition from 2010 to nowadays.

  • End-to-end ASR systems.
  • Convolutional neural networks for speech processing.
  • Recurrent neural networks.
  • Connectionist temporal classification (CTC).
  • Hybrid Transformers with CTC.

Slides

slides.

Practical work

speech commands.

(Back to Main Page)