Introduction to Speech & Natural Language Processing

Lecture 9

Speech Production & Phonetics

​

Krishnendu Ghosh

​

What is Speech?

• Produced from human’s vocal apparatus (Mouth)

​

• Natural mode of communication among humans

​

• Best communication aid for humans

Speech Processing: Applications

• Speech Recognition
• Speaking interface with machines
• Automatic dictation system
• Healthcare
• Speech Synthesis
• Speaking interface with machines
• Voice response system
• Screen readers, reading story books, etc..
• Aid for visually challenged people
• Speaker Recognition/Speaker Verification
• Voice based person authentication system
• Forensic investigation application

Speech Processing: Applications

• Language Identification
• Voice-based Information Retrieval
• Speech Enhancement
• Pathological Speech Analysis: Detection & Classification of Voice Disorders
• Speech Coding
• Paralinguistic Analysis
• Emotion recognition
• Speaker style modeling (speaking rate, pronunciation, etc..)

Developing Speech Applications

• Speech Recognition: Features to characterize sound units
• Speech Synthesis: Parameters of Vocal-tract and Excitation
• Speaker Recognition: Features to characterize speaker
• Language Identification: Features to characterize language
• Information (data) Retrieval (voice-based): Pattern discovery
• Speech Coding: Features for efficient coding and reproduction
• Speech Enhancement: Features to characterize speech, non-speech, noise, reverberation, etc...

Elements of Speech Communication

• Talker: Message formulation and conveying via speech mode.
• Listener: Reception of speech and message comprehension.
• Medium: Physical medium which carries speech from talker to listener.

Speech Chain:

Steps in Human Speech Communication

Speech Production System

• Speech is produced during exhalation of air
• Lungs & associated structure provides required energy
• Vocal-folds inside larynx is the main excitation source and constriction (total or partial) along vocal tract is an additional source
• Supra-glottal system which includes pharynx, oral cavity and nasal cavity behave as time-varying resonator

Excitation Sources

Voiced Excitation

• Vibration of vocal folds
• Voiced speech

Unvoiced Excitation

• Total constriction along the vocal tract
• Partial constriction along the vocal tract
• Unvoiced speech

Mixed Excitation

• Combination of above
• Mixed speech

Production of Speech Sounds

Vowels

• Articulation: Oral cavity is wide open; no major obstruction. Tongue position and lip shape determine vowel type. Vocal folds vibrate (voiced).
• Example Sounds: /a/, /e/, /i/, /o/, /u/
• Example Words: cat, bed, sit, dog, put

Unvoiced Consonants

• Articulation: There’s complete constriction or closure somewhere in the vocal tract, and no glottal vibration.
• Example Sounds: /p/, /t/, /k/, /s/, /f/
• Example Words: pen, top, cat, sun, fan

Production of Speech Sounds

Voiced Consonants

• Articulation: Constriction like unvoiced consonants, with glottal vibration.
• Example Sounds: /b/, /d/, /g/, /z/, /v/
• Example Words: bat, dog, go, zoo, van

Nasal Sounds

• Articulation: Oral cavity closed, nasal cavity opened, air pass through nose.
• Example Sounds: /m/, /n/, /ŋ/ (as in sing)
• Example Words: man, nose, song

Fricatives

• Articulation: Partial closure creates a narrow opening, causing continuous friction as air passes through.
• Example Sounds: /f/, /v/, /s/, /z/, /ʃ/ (as in shoe), /ʒ/ (as in measure)
• Example Words: fan, vase, sun, zebra, shoe, measure

Graphical Model: Speech Production

Digital Model: Speech Production

Speech Perception System

Speech Perception Mechanism

Mainly 3 regions: outer ear, middle ear & inner ear

• Outer ear directs speech pressure variations towards the middle ear
• Middle ear transforms pressure variations into mechanical motion
• Inner ear converts mechanical vibrations into electrical firings in the
• auditory neurons, which leads to brain
• Language decoding and message understanding at the higher centers of learning which is less understood

Steps in Speech Reception and

Message Comprehension

• Acoustic pressure variations funnelled into middle ear by outer ear.
• Eardrum converts acoustic pressure variations to mechanical vibrations.
• Mechanical vibrations are transferred to inner ear by middle ear bones.
• Standing wave patterns are generated in inner ear liquid.
• Standing waves are converted into neural firings on auditory nerve.
• Neural firings are decoded and message comprehension is done in brain.