Moved much of this to Drafts instead

• phonbank: poor articulation

• disfluent kids

• late talkers

• Write a review about ASR benchmark methods

  • REV would be our benchmark
    • What corpora we use?
    • Has anyone used disordered speech?
    • Or really seriously accented speech vis a vi CORALL (how was CORALL sampled?)
  • What samples? How do we sample? What are the benchmarks?

• ASR model + WER

• tildes and noprompt swapped

• WER

• missing words

Lit Survey

Pipeline

Segmentation

Step 0: know what you are building.

breaking tasks

The process of breaking tasks down.

• We need to research tasks to see how complex they are + how to break them down
• Research takes time! It should be its own task
• Over the process of research, the task becomes much simpler

estimating tasks

Requirement: tasks should always be estimated by the person doing the work.

• Task Estimation should be done each time! tasks shift
• Estimate only in powers of 2: 30 minutes, 1h, 2h, 4h, 8h, etc.
• If you never done something before, double the time than you estimate
• If you are teaching someone to do something, quadruple the time than you estimate
• Add buffer time (*1.5), especially if you think yourself as a procrastinator
  • Focus is draining! You need breaks. Take breaks.
  • Things will go wrong! Plan for it.

time iterating

• If anything is longer than 8 hours, that’s a good sign you need to break it down!
• Likely that you have to break things down

MVP

You probably don’t have time to build your feature list

The taxicab norm is a norm against a gridded system; it should follow the same properties of the norm, but not inner products.