S04 - Speech Recognition (GSLC)

What is the main purpose of wav2vec models?

• Text-to-Speech (TTS) conversion

• Self-supervised learning for speech recognition

• Machine Translation

• Image classification

What is the primary difference between wav2vec and wav2vec 2.0?

• wav2vec 2.0 uses CNNs, while wav2vec uses Transformers

• wav2vec 2.0 removes dependency on phoneme-based features

• wav2vec is multilingual, while wav2vec 2.0 is not

• wav2vec 2.0 is only used for speaker recognition

Which component of wav2vec 2.0 is responsible for capturing long-range speech dependencies?

• Feature Encoder

• Transformer Context Network

• Quantization Module

• Spectrogram Processor

What is the function of the Quantization Module in wav2vec 2.0?

• Converts audio into text

• Discretizes speech representations into codebook embeddings

• Removes noise from speech signals

• Translates speech into multiple languages

What is the main advantage of XLSR over wav2vec 2.0?

• It is trained on multiple languages

• It does not use self-supervised learning

• It does not require any training

• It is only used for speaker verification

What does the “53” in XLSR-53 refer to?

• The number of Transformer layers

• The number of audio features extracted

• The number of languages used for training

• The number of training datasets

Which model is best suited for low-resource languages?

• wav2vec

• wav2vec 2.0

• XLSR-53

• X-LSR

What technique does wav2vec 2.0 use for self-supervised learning?

• Spectrogram Processing

• Contrastive Learning

• Masked Predictive Coding (MPC)

• Recurrent Neural Networks (RNNs)

How does the Feature Encoder in wav2vec 2.0 process raw speech?

• Uses 1D Convolutional Layers to extract low-level features

• Converts speech into phonemes

• Uses spectrograms for processing

• Directly outputs text

What is the main benefit of X-LSR compared to XLSR-53?

• X-LSR is trained on 128 languages, offering better multilingual performance

• X-LSR is only used for English ASR

• X-LSR removes the need for transformers

• X-LSR does not support fine-tuning