In this module’s lab, you continue working on your ASR assignment.

This week you should aim to get your speaker independent digit recognition pipeline set up.  This means writing scripts so that you can:

• Train your digit recogniser on speech data from multiple speakers in the course dataset
• Apply the digit recogniser to test set data from multiple speakers in the dataset.
  • Make sure you can handle the case where the training set speakers and test set speakers are completely different people.
• Collate and evaluate the results of the digit recogniser in terms of Word Error Rate

Once you’ve got your speaker independent pipeline going, you should think about what experiments you want to focus on for the report write up.  This means identifying some hypotheses and explaining why you think they will (or won’t) hold in specific experimental settings.  You should support your arguments with sources from the literature (i.e., readings from the course and beyond).

Discuss potential hypotheses and experimental designs in with Simon and the tutors in the lab.