I. Introduction to the physiology of speech and hearing

Rabiner & Juang, Chapter 1, 2.1 - 2.2

Speech production physiology

Figure: The human vocal system.
After [J. L. Flanagan, Speech Analysis and Perception, Springer-Verlag, Berlin, 2nd edition, 1965].

Figure: Schematic of the functional components of the vocal system.
After [J. L. Flanagan, Speech Analysis and Perception, Springer-Verlag, Berlin, 2nd edition, 1965].

The process of producing speech sounds:

• lungs: fill with air
• contraction of rib cage forces air from the lungs into the trachea - the volume of air determines the amplitude of the sound
• trachea (windpipe): conveys air to the vocal tract. The vocal cords, at the top of the trachea, separate the trachea from the base of the vocal tract
• vocal tract
  • consists of:

    pharynx (throat)

    mouth

    nose

  • the shape and size of the vocal tract vary by positioning the articulators: the tongue, teeth and lips
  • the shape of the vocal tract determines the type of speech sound - e.g., the /a/ in "hat" vs the /i/ in "hit"

Speech differs from breathing in that at some point in the path you set the air in rapid motion or vibration

Two principal components of speech production

1. Excitation - create a sound by setting the air in rapid motion
2. Vocal tract - "shape" the sound

A. Excitation: three principal forms

1. Phonation: vibration of vocal cords

   The vocal cords consist of ligament and muscle, and are adjustable under muscle control. The cartilage surrounding the vocal cords provides support. The opening that allows air to pass through the vocal cords from the trachea to the larynx is called the glottis.

   There are two modes of operation of the vocal cords:

   1. Vibrating
      • cords tense, pressed together - no air flows
      • air pressure from the lungs forces them open
      • local pressure is reduced --> cords close
      • the cycle repeats

      The result is a quasi-periodic release of air into the pharynx. The fundamental frequency of the vocal cord opening/closing cycle becomes the fundamental frequency (informally, the "pitch") of the resulting sound.

      The tenser the vocal cords
      

      -- the higher the pitch

      -- the shorter the period

      Typical frequency of vocal cord open/close cycle:

      male: 128 Hz

      female: 256 Hz

   2. Non-vibrating
      • vocal cords open
      • air flows from trachea to pharynx without interruption
      • not an excitation since the air isn't set into rapid motion

2. Frication: Turbulent air flow
   • "noise source"
   • the excitation is set up by forcing air past a constriction at some point in the vocal tract

     e.g., /f/ in "for": top teeth & bottom lip

     e.g., /th/ in "thin": tongue & top teeth

   • can combine frication with phonation

     e.g., /v/ as in "vote" top teeth & bottom lip as in /f/, combined with phonation

   • one special case: form constriction by partially closing the vocal cords, held rigid - not periodic airflow

     get /h/ as in hot, or whispering

3. Plosive: Closure at some point in the vocal tract, followed by a release of air
   e.g., /p/ as in "pot": closure at lips

   • can be combined with vocal cord vibration: phonation and plosive

     /b/ as in "boy": closure at lips closure as in /p/, combined with phonation

A model of speech production

Relation between excitation and the vocal tract

Simple model:

Linear system, with input x(t), system function H[], and output s(t)

• Time invariant? no, but maybe "piecewise" time invariant - e.g. every 10 ms.
• h(t) impulse response
• s(t) = x(t) * h(t)

1. Excitation - create a sound by setting the air in rapid motion
2. Vocal tract - "shape" the sound

Reference: Rabiner & Schafer 3.2, 3.3

Simple model: uniform tube closed at glottis end, open at mouth end

/a/ as in "ado" - called "schwa"

Modes of vibration: resonances

The uniform tube has resonant frequencies at 500, 1500, and 2500 Hz.

These are the same resonances that characterize the "schwa" sound.

In speech, resonant frequencies are called formant frequencies.

Can show that the vocal tract modeled as a uniform tube can be represented with an all-pole transfer function.

In terms of the LTI system model, the excitation is the input function x(t), the vocal tract acts as the system function H[], and the speech is the output s(t).

The shape of the resulting spectrum is given by H[f]. In a simplified model for phonated sounds, the glottal pulses of x(t) form an impulse train, with interval T between pulses. This appears in the speech spectrum as pulses at frequency intervals 1/T, shaped by the H[f] envelope.

In reality, the vocal system differs from the uniform tube model in several respects:

• In the vocal tract:
  • losses through walls
  • reflections from lips
  • radiation from mouth
• In the excitation:
  • actual shape of glottal pulses

Model breakdowns:

• Nasals: uniform tube with side branch
  
  H(f) has zeroes - typically one zero in 1-2 kHz range

  Fricatives, plosives: posterior and anterior tubes, with source (excitation) between them
  introduces zeros. There may be no resonances, or resonances may be hidden by zeros.

Speech analysis systems typically assume all-pole filters are identical to tube models for all speech sounds.

Auditory system: hearing and perception

Reading: Rabiner & Juang section 3.5

Figure: Schematic of the human ear (not to scale).
After [J. L. Flanagan, Speech Analysis and Perception, Springer-Verlag, Berlin, 2nd edition, 1965].

Outer ear: collects sound waves

• amplifies sound waves in some frequencies
• vibrations of air are translated to vibrations of the tympanic membrane

• performed by the ossicles
• amplification 15:1

Inner ear:

• cochlea
• nerve endings

Figure: Schematic of the organ of Corti, in inner ear.

After [J. L. Flanagan, Speech Analysis and Perception, Springer-Verlag, Berlin, 2nd edition, 1965].