WorksheetsSpeech Recognition: Mid Test Preparation
Total questions: 45
Worksheet time: 16mins
What is the primary goal of speech recognition?
To analyze the emotional tone of speech
To convert spoken language into text
To generate human-like voices
To translate languages in real time
Which of the following is NOT a challenge for automatic speech recognition?
Diverse dialects
Homophones
Perfect phonetic transcription
Background noise
What is the role of phonetics in speech recognition?
To identify the speaker's emotions
To model the structure of language
To analyze how speech sounds are produced and perceived
To improve speech synthesis
What is an acoustic model in speech recognition?
A model that translates text into speech
A model that maps audio signals to phonemes
A model that detects emotional tone in speech
A model that transcribes handwritten text
What type of speech recognition where the model can recognize speech of an individual even though his/her speech doesn't exist in the training set to build the model?
Speaker-dependent
Speaker-independent
Isolated-word recognition
Continuous speech recognition
Speaker verification is the same as speaker recognition.
True
False
Speaker recognition is a multiclass classification.
True
False
What is the purpose of speech synthesis (TTS)?
To transcribe speech into text
To generate human-like speech from text
To classify different speakers
To remove noise from speech
Why is human-level speech recognition still difficult for machines?
Computers cannot process large datasets
Humans do not use grammar in speech
Machine learning cannot be applied to speech
Speech varies in accents, intonation, and background noise
What are common errors in automated speech recognition systems?
Homophone confusion
Mispronunciation
Inability to recognize long sentences
Perfect recognition every time
Which of the following is NOT an application of speech recognition?
Voice assistants
Medical transcription
Stock market prediction
Automated customer service
Which of the following is an example of speaker verification?
Identifying the language spoken
Converting spoken words into text
Determining if a specific person spoke a sentence
Translating speech into another language
What are the three main components of speech-to-speech translation?
Speech recognition, machine translation, speech synthesis
Phonetics, grammar, vocabulary
Audio encoding, speech compression, speech enhancement
Text processing, data mining, speech transcription
Which of these factors most affects the accuracy of speech-to-text conversion?
The language of the speaker
The amount of labeled training data
The font size of the text
The color of the speech interface
What is the primary advantage of end-to-end neural network approaches in speech recognition?
They require no labeled data
They use only rule-based approaches
They eliminate the need for speech synthesis
They simplify the traditional pipeline of speech recognition
What is phonetics?
The study of word meanings
The study of how speech sounds are produced and perceived
The study of sentence structures
The study of written language
What is the International Phonetic Alphabet (IPA) used for?
Translating languages
Representing the sounds of speech in a consistent way
Writing phonetic poetry
Encoding Morse code
What is ARPAbet?
A system used for computer programming
A method for speech synthesis
A phonetic transcription system for American English
A set of musical notes
Which of the following is NOT a place of articulation?
Labial
Alveolar
Velar
Rhotic
Which of the following is an example of a fricative?
/ p /
/ t /
/ f /
/ m /
What is prosody in speech?
The study of individual speech sounds
The rhythm, stress, and intonation of speech
The process of word formation
The ability to recognize spoken words
What is a spectrogram used for?
Measuring sentence length
Visualizing the frequency components of speech over time
Translating spoken language into text
Determining the grammatical structure of a sentence
How does an accent affect speech recognition?
It changes the pronunciation of phonemes, making recognition more difficult
It does not affect speech recognition at all
It only affects high-pitched voices
It is only an issue for text-to-speech systems
Why are modern speech recognition systems less reliant on phonetic transcription?
They use end-to-end deep learning approaches that directly learn from raw speech data
They no longer need to process speech sounds
Phonetics is only useful for human linguists
They rely solely on dictionary-based word matching
Which model was traditionally used for acoustic modeling in speech recognition?
Deep Neural Networks (DNNs)
Support Vector Machines (SVMs)
Gaussian Mixture Models (GMMs) with Hidden Markov Models (HMMs)
Decision Trees
In an HMM-GMM system, what does the Gaussian Mixture Model (GMM) estimate?
The probability of phoneme sequences
The probability of observing an acoustic feature given a hidden state
The probability of a word occurring in a sentence
The correct grammatical structure of a sentence
What is a triphone in acoustic modeling?
A type of vowel sound
A three-layer deep neural network
A phoneme model that considers preceding and following phonemes
A type of microphone used in speech recognition
What problem does state tying solve in triphone modeling?
The need for too many phoneme labels
The difficulty of detecting consonants
The low accuracy of vowel recognition
The excessive number of triphone states in training data
What is the Mel-Frequency Cepstral Coefficient (MFCC) primarily used for?
Speech synthesis
Speech signal feature extraction
Grammar correction
Machine translation
What is the purpose of Mel-Scale filtering in MFCC extraction?
To enhance high-frequency components
To remove noise from speech signals
To simulate how human hearing perceives different frequencies
To convert speech into text
Which transformation is used to convert speech signal from time domain to frequency domain in extracting MFCC?
Discrete Cosine Transform
Inverse Discrete Fourier Transform
Hamming Windowing
Fast Fourier Transform
Which technique is commonly used to improve speech recognition on low-resource languages?
Transfer learning from high-resource languages
Ignoring phonetic variations
Only training models on written text
Removing vowels from speech data
Which of the following components is NOT typically found in an ASR system?
Acoustic Model
Lexicon
Image Processing Module
Language Model
What is the role of the acoustic model in an ASR system?
To predict the meaning of words
To map speech features to phonemes or subphonetic units
To convert speech into images
To generate speech from text
What is the function of the language model in ASR?
To predict the likelihood of a sequence of words
To analyze the acoustic features of speech
To extract features from raw speech data
To convert phonemes into waveforms
In an HMM used for speech recognition, what do the hidden states represent?
Words in a sentence
Acoustic feature vectors
Phonemes or subphonetic units
The noise level in an audio signal
What is Word Error Rate (WER) in ASR evaluation?
A measure of speech synthesis quality
The amount of noise present in a speech signal
The speed at which an ASR system transcribes speech
The percentage of incorrectly recognized words in ASR output
Which feature extraction method is most commonly used in ASR?
MFCC (Mel-Frequency Cepstral Coefficients)
Principal Component Analysis (PCA)
Spectral Clustering
Fourier Series Expansion
What is the main advantage of Connectionist Temporal Classification (CTC) in speech recognition?
It eliminates the need for word dictionaries
It allows training without explicit alignment of input and output sequences
It only works for single-word speech recognition
It requires fewer training examples than HMMs
Connectionist Temporal Classification (CTC) recognizes a speech utterance in the unit of phones.
True
False
There are two steps in CTC. What are they?
extracting
searching
mapping
compressing
CTC requires lexicon during model training.
True
False
Is it possible to recognize Out of Vocabulary (OOV) words when using CTC for speech recognition?
yes
no
Listen, Attend, Spell (LAS) has two components. They are _______
mapper
reducer
encoder
decoder
The epsilon symbol exists in the output of CTC, not LAS.
True
False
