Kaldi Speech Recognition Toolkit Quiz

Answers and explanations appear as you take the quiz.

1. Which programming language is used for Kaldi’s documented core implementation?
   • Java
   • Rust
   • Python only
   • C++
2. What mathematical structure does Kaldi use to combine its acoustic model, lexicon, and language model into one decoding graph?
   • Recurrent neural networks
   • Support vector machines
   • Weighted finite-state transducers (WFSTs)
   • Hidden Markov trees
3. Who is listed first in the project’s recommended 2011 Kaldi toolkit paper citation?
   • Geoffrey Hinton
   • Daniel Povey
   • Yann LeCun
   • Andrew Ng
4. In a classic Kaldi GMM acoustic model, what do Gaussian mixtures model?
   • Acoustic likelihoods for speech states
   • Only the spelling of every output word
   • Only a file’s speaker-ID string
   • The application’s interface color
5. Which related project describes FSA/FST algorithms with differentiation and PyTorch compatibility?
   • A universal replacement requirement for every Kaldi recipe
   • Only the original HTK text formatter
   • k2
   • An image-compression file format
6. What does the 'L' represent in Kaldi's HCLG decoding graph?
   • The loudness normalization
   • The language model
   • The loss function
   • The pronunciation lexicon
7. Which license does the official Kaldi project documentation list?
   • Apache License version 2.0
   • A proprietary-only research license
   • An exclusive GPL version 3 requirement
   • No license terms of any kind
8. What association is recorded by a Kaldi data-directory utt2spk row?
   • A word mapped to its pronunciation phones
   • An utterance identifier mapped to a speaker identifier
   • A decoding state mapped to a language-model probability
   • A feature matrix mapped only to its pixel dimensions