Listen Attend and Spell Quiz

Answers and explanations appear as you take the quiz.

1. What are the two main components of the LAS model?
   • Listener and Speller
   • Encoder and tokenizer
   • Generator and discriminator
   • Filter and classifier
2. What does the Speller in LAS output?
   • Phonemes from a dictionary
   • Characters, one at a time
   • Whole words only
   • Frequency spectra
3. Why is the LAS encoder called 'pyramidal'?
   • It is shaped like a triangle on screen
   • It uses three GPUs
   • It reduces the time resolution at each layer
   • It pyramids the loss values
4. What older component does LAS notably NOT require?
   • A microphone
   • A loss function
   • Training data
   • A hand-built phoneme pronunciation dictionary
5. Which mechanism lets the Speller focus on the relevant part of the audio for each character?
   • Attention
   • Dropout
   • Batch normalization
   • Beam pruning
6. What is one limitation of the original LAS model?
   • It could not run on any hardware
   • It was non-streaming, needing the full utterance before decoding
   • It only output numbers
   • It required a phoneme dictionary
7. What WER did the LAS paper report on its Google voice-search subset without a language model?
   • 1.41%
   • 41.0%
   • 14.1%
   • 80.0%
8. What distinction does the LAS paper make about dependencies among output characters?
   • It predicts each character independently with no sequence model
   • It outputs only phoneme classes and forbids character context
   • It requires a separate dictionary to define character dependencies
   • The model emits character sequences without assuming the characters are independent