Text and speech processing

1




Important Questions and Answers
CP25001-Text and Speech Processing
Syllabus
1. Introduction to Text and Speech Processing: NLP pipeline overview Human speech
production and acoustic phonetics Text normalization: tokenization, stemming,
lemmatization Basics of speech signal: sampling, frequency, noise.
Activities:
• Speech waveform analysis using Praat
• Tokenization and stemming with NLTK .
2. Text Processing Techniques: Part-of-speech tagging Parsing techniques: CFG, Dependency
Parsing Named Entity Recognition (NER), Chunking, Language modeling: N-gram models.
Activities:
• POS tagging using spaCy
• Building a rule-based NER system.
3. Speech Signal Processing: Feature extraction: MFCC, PLP, LPC Acoustic modeling:
HMM, GMM Basics of ASR (Automatic Speech Recognition) Hidden Markov Models for
speech.
Activities:
• Extract MFCC features using Python Group
• Basic ASR using pre-trained models.
4. Speech Synthesis and Applications: Text-to-speech (TTS) systems Speech databases and
corpora Evaluation metrics: WER, CER Applications: Voice Assistants, Dictation Systems.
Activities:
• TTS Demo using pyttsx3
V.




• Compare open-source TTS tools.
U
m




5. NLP and Speech in Practice: Sentiment analysis Chatbots and dialog systems Multilingual
ap




and low-resource NLP Ethics in speech and text technology.
at




Activities:
hy




• Building a chatbot using Rasa
,Pr




• Ethical issues in speech surveillance.
of
./
EC
N
ot
es

, 2


6. Models for NLP: RNN’s and LSTM – The Transformer – Large Language Models –
Masked Language Models.
Activities: Create a debate to Compare and contrast the pros and cons of various models for
NLP.
S.No. Description
Unit-I
Introduction to Text and Speech Processing
Basic Concepts
Concept Important Basic Points
Natural Language NLP is a branch of Artificial Intelligence that enables computers to
Processing (NLP) understand, process and generate human languages.
Main stages: Text collection → Text cleaning → Tokenization →
NLP Pipeline
Normalization → Feature extraction → Model processing → Output.
Converts raw text into a structured format for computer analysis. It includes
Text Processing
cleaning, removing stop words and handling punctuation.
Human Speech Speech is produced by lungs (air source), vocal cords (sound source), and
Production vocal tract (sound shaping).
Lungs, trachea, larynx, vocal cords, tongue, teeth, lips, palate and nasal
Speech Organs
cavity.
Study of physical properties of speech sounds, such as frequency, amplitude,
Acoustic Phonetics
duration and pitch.
Smallest unit of sound that distinguishes meaning. Example: /p/ and /b/ in
Phoneme
pat and bat.
Converts text into a standard format by removing unnecessary symbols,
Text Normalization
converting case and expanding abbreviations.
Splitting text into smaller units called tokens (words, sentences or
Tokenization
subwords). Example: "I love NLP" → I, love, NLP.
Removes prefixes or suffixes to obtain the root-like form of a word.
Stemming
Example: playing → play.
Converts a word into its meaningful dictionary base form. Example: better
Lemmatization
→ good; running → run.
Speech is an analog signal produced by human vocal organs and converted
V.




Speech Signal
into digital form for computer processing.
U
m




Process of converting a continuous-time signal into discrete-time samples at
ap




Sampling
regular intervals.
at
hy




Number of samples taken per second, measured in Hertz (Hz). Example:
Sampling Frequency
,




16,000 Hz.
Pr




Frequency Number of cycles of a sound wave per second. It determines the pitch of
of
./
EC
N
ot
es

, 3


sound and is measured in Hz.
Amplitude Maximum strength or magnitude of a sound wave. It is related to loudness.
Unwanted sound or interference that affects speech quality and recognition
Noise
accuracy.
Process of removing background noise using filtering and signal-processing
Noise Reduction
techniques.
Converts spoken language into written text using acoustic and language
Speech Recognition
models.
Important Equations (Copy to Word)

1. Sampling Frequency:

Fs = 1 / Ts

2. Sampling Theorem (Nyquist):

Fs ≥ 2 × Fmax

3. Time Period and Frequency:

T=1/f

4. Digital Speech Signal:

x[n] = x(nTs)

5. Frequency:

f = Number of cycles / Time

6. Signal-to-Noise Ratio (SNR):
V.




SNR (dB) = 10 log10 (Signal Power / Noise Power)
U
m




7. Number of Samples:
ap
at
hy




N = Fs × Duration
, Pr
of
./
EC
N
ot
es

, 4



Important Differences for Exams

Lemmatization
Stemming
Removes prefixes or suffixes. Finds dictionary base form.
May produce meaningless words. Produces meaningful words.
Faster processing. Requires linguistic knowledge.
Example: studies → studi. Example: studies → study.
Exam focus

Important topics to study: NLP pipeline, speech production organs, acoustic phonetics, tokenization,
stemming vs lemmatization, sampling theorem, frequency and noise.

Remember: Sampling theorem and SNR formulas are important for numerical problems in text and
speech processing.
Multiple Choice QA
1. Which of the following is the first major stage in the NLP pipeline?

A) Speech recognition
B) Text preprocessing
C) Text generation
D) Machine translation

Answer: B) Text preprocessing

Reason: Text preprocessing is an important initial stage in NLP. It involves cleaning, tokenization,
normalization and removing unwanted characters before further analysis.

2. Which part of the human body is primarily responsible for producing voiced speech sounds?

A) Tongue
B) Lips
V.




C) Vocal cords
U




D) Teeth
m
ap




Answer: C) Vocal cords
at
hy
,




Reason: Voiced sounds are produced when air from the lungs causes the vocal cords in the larynx to
Pr




vibrate. The tongue, lips and teeth help shape speech sounds.
of
./
EC
N
ot
es