Short Answer

Limits of Hand-Built Speech Features

Question: Give two reasons a speech system can be held back by manually designed features rather than learning from the raw signal.

Sample answer: One limitation is that the preprocessing step may compress the audio so much that useful detail is lost before the model sees it. Another is that the system may be forced to work through a human-chosen intermediate code, such as a fixed set of sound categories, even if that is not the best representation for the task.

Key points:

  • Aggressive preprocessing can remove subtle but important information.
  • A predefined intermediate representation can create a bottleneck that the model cannot improve on.

Rubric: Full credit requires both the loss of useful detail and the restriction caused by an imposed intermediate representation.

0

1

Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI