Skip to main content

Speech recognition and graph transformer nets

Difficulty level

This lecture provides an introduction to the problem of speech recognition using neural models, emphasizing the CTC loss for training and inference when input and output sequences are of different lengths. It also covers the concept of beam search for use during inference, and how that procedure may be modeled at training time using a Graph Transformer Network. It is a part of the Deep Learning Course at CDS, a course that covered the latest techniques in deep learning and representation learning, focusing on supervised and unsupervised deep learning, embedding methods, metric learning, convolutional and recurrent nets, with applications to computer vision, natural language understanding, and speech recognition. Prerequisites for this module include: Modules 1 - 5 of this course and Introduction to Data Science or a Graduate Level Machine Learning.

Topics covered in this lesson


00:00 – Guest lecturer introduction
01:10 – Outline
02:36 – Modern speech recognition
09:26 – Connectionist temporal classification
54:44 – Decoding with beam search (inference)
1:11:09 – Graph Transformer Networks