ASR: Labs

Compulsory weekly lab sessions run in weeks 3-7, in which you will use OpenFst and Python to build your own small-vocabulary speech recognition system from scratch. All labs will run in Appleton Tower in room 5.05. You need only attend one lab session per week – you should already have been allocated one of the sessions.

  • Lab group 01: Tuesdays 14.10
  • Lab group 02: Tuesdays 15.10
  • Lab group 03: Wednesdays 10.00

Labs are accessed via a GitHub repository with new exercises added at the start of each week. Setup instructions are contained in the README file.

An introduction to the labs is presented in Lecture 5 (slides). You will work in pairs, with each pair submitting one set of solutions per lab. The five labs together are worth 10% of the total mark for the course; each set of exercises will be assessed by your demonstrator. To be awarded the marks, you will need to show the demonstrator your solutions by the end of the following week's lab – both members of your pair should be present for this.

Lab topics

  • Lab 1 (Week 3): Building HMMs with OpenFst
  • Lab 2 (Week 4): Computing HMM probabilities
  • Lab 3 (Week 5): Viterbi decoding (1)
  • Lab 4 (Week 6): Viterbi decoding (2)
  • Lab 5 (Week 7): WFST operations
     

Notes: no labs run during flexible learning week. Labs 3 and 4 share one Jupyter notebook.  

 

This page maintained by Peter Bell.

License
Creative Commons - Attribution Non-Commercial Share A Like