S03 - Speech Recognition (S1)

What is the primary function of an acoustic model in speech recognition?

• Convert text into speech

• Match audio signals to phonemes

• Remove background noise

• Translate languages

Which of the following models is traditionally used for acoustic modeling?

• HMM-GMM

• CNN-RNN

• Transformer

• LSTM-CRF

In Gaussian Mixture Models (GMMs), what is the purpose of mixture weights?

• Determine the probability of each Gaussian component

• Define the number of states in HMM

• Perform backpropagation

• Reduce dimensionality

What is a triphone in the context of phonetic modeling?

• A single phoneme

• A three-letter word

• A phone with context from neighboring phones

• A type of speech corpus

What is the main advantage of state tying in acoustic modeling?

• Reduces the number of unique triphones required

• Increases computational complexity

• Eliminates phoneme variability

• Converts speech to text

Which feature extraction method is most commonly used in speech recognition?

• Fourier Transform

• Mel-Frequency Cepstral Coefficients (MFCC)

• Wavelet Transform

• Linear Regression

What is the main reason MFCC is widely used in speech recognition?

• It is computationally expensive

• It is resistant to background noise

• It mimics human auditory perception

• It increases data redundancy

Which adaptation technique is best suited for speaker adaptation with small amounts of data?

• Maximum Likelihood Linear Regression (MLLR)

• Maximum A Posteriori (MAP) Adaptation

• Convolutional Neural Networks

• Hidden Markov Models

Why are deep neural networks (DNNs) preferred over traditional HMM-GMM models?

• They require less data for training

• They improve feature learning and generalization

• They are less complex than GMMs

• They only use context-independent phonemes

What is the purpose of SpecAugment in training DNN models for speech recognition?

• Convert speech to text 

• Introduce variations in the data to improve generalization 

• Remove background noise

• Reduce training time