Glottal Source Power and Inverse Filtering

Measuring the acoustic power generated at the glottal source requires separating the effects of glottal airflow from vocal tract filtering. Inverse filtering provides the primary technique for recovering the glottal flow waveform from radiated speech pressure, enabling quantification of source power and analysis of how laryngeal adjustments affect acoustic output.

The Source-Filter Framework

Voice production involves two relatively independent subsystems that can be analyzed separately.

Fundamental Principle

Source-Filter Theory

Source: glottal airflow generates acoustic excitation
Filter: vocal tract modifies source spectrum
Independence: source and filter approximately decoupled
Linearity: system behaves approximately linearly
Superposition: output = source × filter transfer function

Mathematical Representation

In the frequency domain:

P_rad(ω) = U_g(ω) × H_tract(ω) × H_rad(ω)

Where:

P_rad(ω): radiated pressure spectrum
U_g(ω): glottal flow spectrum (source)
H_tract(ω): vocal tract transfer function (filter)
H_rad(ω): radiation transfer function

Advantages of Separation

Isolates laryngeal function from articulatory effects
Enables measurement of pure source characteristics
Facilitates comparison across vowels and speakers
Clarifies mechanisms of intensity control
Essential for voice analysis and synthesis

Limitations of Direct Measurement

Invasive Techniques

Direct glottal airflow measurement requires transglottal sensors
Electroglottography (EGG) provides contact area, not flow
High-speed videoendoscopy shows motion, not airflow
Particle image velocimetry research-only, not clinical
Practical clinical measurement requires indirect methods

Radiated Speech Limitations

Microphone captures combined source-filter-radiation output
Cannot directly observe glottal waveform
Vocal tract formants obscure source characteristics
Different vowels produce different spectra from same source
Requires computational separation of components

Inverse Filtering Technique

Inverse filtering computationally removes vocal tract and radiation effects to recover glottal flow.

Basic Concept

Forward Process (Voice Production)

Glottal Flow → Vocal Tract → Radiation → Radiated Pressure

Inverse Process (Analysis)

Radiated Pressure → Inverse Radiation → Inverse Vocal Tract → Glottal Flow

Mathematical Formulation

In the frequency domain:

U_g(ω) = P_rad(ω) / [H_tract(ω) × H_rad(ω)]

Or equivalently:

U_g(ω) = P_rad(ω) × H_tract⁻¹(ω) × H_rad⁻¹(ω)

Where H⁻¹ denotes the inverse transfer function.

Implementation Steps

Inverse filtering process Figure 9.4: Block diagram of inverse filtering showing sequential removal of radiation and vocal tract effects to recover glottal flow from speech pressure signal.

Step 1: Remove Radiation Effect

The radiation characteristic (+6 dB/octave) is removed by integration:

Time-domain: integrate the pressure signal
Frequency-domain: divide by jω
Result: volume velocity at lips
Simple first-order operation

Step 2: Estimate Vocal Tract Transfer Function

Several methods exist:

Linear prediction coding (LPC): assumes all-pole model
Formant tracking: identifies resonance frequencies
Cepstral analysis: separates source and filter
Manual formant selection: interactive adjustment

Step 3: Remove Vocal Tract Filtering

Apply inverse of estimated vocal tract transfer function:

Cancels formant peaks
Removes anti-resonances
Result: glottal flow waveform
Sensitive to accurate formant estimation

Step 4: Verify and Refine

Check recovered waveform for:

Realistic pulse shape
Appropriate open quotient
Smooth closure
Minimal residual formant structure
Iterate if necessary

Challenges and Sources of Error

Vocal Tract Estimation

LPC order must be appropriate (typically 10-14 coefficients)
Subglottal resonances can interfere
Nasal coupling complicates model
Time-varying tract during transients
Individual anatomical variation

Source-Tract Coupling

Assumption of independence not perfect
Source impedance affects tract resonances
Nonlinear effects at high amplitudes
Particularly important for high-intensity phonation
May require iterative refinement

Radiation Model Assumptions

Simple integration assumes ideal point source
Head/torso effects not captured
Directivity ignored
Adequate for most clinical purposes
More sophisticated models available for research

Signal Quality Requirements

High signal-to-noise ratio needed
Calibrated microphone essential
Steady phonation simplifies analysis
Running speech more challenging
Background noise problematic

Glottal Flow Waveform Characteristics

The recovered glottal flow reveals important source features.

Temporal Parameters

Open Phase Duration (T_o)

Time when glottis open and airflow occurs
Typically 40-70% of period in modal voice
Related to open quotient Q_o = T_o/T
Varies with adduction and pressure

Flow Rise Time (T_p)

Time for flow to increase from onset to peak
Usually 20-40% of period
Gradual increase during opening

Flow Decline Time (T_n)

Time for flow to decrease from peak to closure
Usually 20-30% of period
More rapid than rise time (skewing)
Abruptness related to collision forces

Peak Flow Amplitude (U_peak)

Maximum volume velocity during cycle
Typically 200-800 cm³/s for conversational speech
Scales with lung pressure and glottal opening
Major determinant of source power

Derived Parameters

Skewing Quotient (Q_s)

Q_s = T_p / T_n

Values > 1.0 indicate skewed waveform
Typical range: 1.0 to 2.0
Related to closure abruptness
Affects spectral slope

Maximum Flow Declination Rate (MFDR)

Peak negative slope during closure
Measured in liters/second/second
Typical range: 200-2000 L/s/s
Strong indicator of collision forces
Correlates with high-frequency spectral energy

AC Flow Component

Time-varying portion of flow
Peak-to-peak amplitude
Determines acoustic power
Distinct from mean (DC) flow

Calculating Glottal Source Power

Power in the glottal source can be quantified from the recovered flow waveform.

Time-Domain Power Calculation

Instantaneous Power

At each time instant, power flow into the vocal tract:

P(t) = p_g(t) × u_g(t)

Where:

P(t): instantaneous power (watts)
p_g(t): glottal pressure (subglottal pressure when open)
u_g(t): glottal flow (volume velocity)

Average Glottal Source Power

P_avg = (1/T) ∫[0 to T] p_g(t) × u_g(t) dt

Where T is the fundamental period.

Practical Calculation

Requires simultaneous pressure and flow measurement
Pressure approximated by subglottal pressure estimate
Integration over multiple cycles
Typically 1-100 milliwatts for speech

Frequency-Domain Power Calculation

Spectral Power Density

From Parseval’s theorem:

P_avg = Σ|U_g(n)|² × R_tract(n)

Where:

U_g(n): glottal flow amplitude at harmonic n
R_tract(n): real part of tract input impedance
Σ: sum over all harmonics

Spectral Slope Relationship

Glottal flow spectrum typically decreases ~12 dB/octave
Power concentrated in low harmonics
Fundamental and second harmonic carry most power
High harmonics contribute little to total power
Spectral richness distinct from total power

Typical Power Values

Conversational Speech

Glottal source power: 1-10 milliwatts
Radiated acoustic power: 10-100 microwatts
Efficiency: 0.1-1% (very low)
Most power dissipated as heat and viscous losses

Loud Speech/Singing

Glottal source power: 10-100 milliwatts
Radiated acoustic power: 100 microwatts to 1 milliwatt
Efficiency: similar percentage range
Absolute power increases but efficiency similar

Power Distribution

Fundamental frequency: 30-50% of total power
Second harmonic: 20-30% of total power
Third harmonic: 10-15% of total power
Higher harmonics: remaining power
Distribution varies with voice quality

Relationship to Radiated Acoustic Power

The glottal source power exceeds radiated power due to losses.

Power Loss Mechanisms

Viscous Losses in Tissues

Energy dissipated as heat in vocal fold tissue
Damping during oscillation
Typically 50-70% of glottal source power
Increases with oscillation amplitude

Absorption in Vocal Tract Walls

Acoustic energy absorbed by soft tissues
Approximately 10-20% of source power
Frequency-dependent (higher frequencies absorbed more)
Yields tissue typically 20-30% of energy

Radiation Losses

Only 1-10% of glottal source power radiated
Most power lost before radiation
Radiation efficiency low at speech frequencies
Improves at high frequencies but still small fraction

Practical Implications

Voice production extremely inefficient acoustically
Most energy becomes heat
Thermal loading of larynx significant during heavy voice use
Cooling by respiration and blood flow essential
Efficiency considerations guide vocal economy strategies

Transfer Functions

Glottal Power to Radiated Power

P_rad = P_glottal × η_tissue × η_tract × η_radiation

Where η represents efficiency at each stage:

η_tissue: mechanical efficiency (~0.3-0.5)
η_tract: transmission efficiency (~0.8-0.9)
η_radiation: radiation efficiency (~0.01-0.1)
Overall: typically 0.001-0.01 (0.1-1%)

Clinical Applications of Inverse Filtering

Inverse filtering provides valuable diagnostic and therapeutic information.

Voice Disorder Assessment

Glottal Insufficiency

Incomplete closure increases DC flow
Reduced AC flow amplitude
Decreased glottal power
Lower efficiency
Breathy voice quality

Hyperfunction/Pressed Voice

Very short open quotient
High maximum flow declination rate
Large collision forces
Excessive glottal power for intensity achieved
Risk of vocal trauma

Vocal Fold Paralysis

Asymmetric flow waveform
Irregular periodicity
Reduced flow amplitude
Low source power
Poor voice intensity capability

Treatment Monitoring

Objective Measures

Track changes in glottal waveform parameters
Document improvements in closure pattern
Monitor source power efficiency
Compare pre/post therapy
Quantify surgical outcomes

Examples

Improved open quotient following medialization
Reduced MFDR with relaxation techniques
Increased AC flow with vocal function exercises
Better waveform regularity with neural recovery

Research Applications

Inverse filtering enables investigation of voice production mechanisms.

Parametric Studies

Effect of Laryngeal Adjustments

Isolate adduction effects on flow
Examine F0 modulation independent of filtering
Study registration phenomena
Investigate coordination patterns

Pressure-Flow Relationships

Measure flow at varying subglottal pressures
Determine laryngeal flow resistance
Calculate phonation threshold
Model aerodynamic-mechanical coupling

Voice Synthesis and Modeling

Model Validation

Compare synthetic flow waveforms with measured
Refine model parameters
Test theoretical predictions
Develop improved voice simulators

Practical Applications

Speech synthesis systems
Voice modification algorithms
Singing voice synthesis
Clinical training simulations

Summary

Glottal source power quantifies the acoustic energy generated by vocal fold vibration, requiring inverse filtering techniques to separate source characteristics from vocal tract filtering and radiation effects. The inverse filtering process sequentially removes radiation (by integration) and vocal tract filtering (using LPC or formant tracking) to recover the glottal flow waveform from radiated speech pressure. The recovered waveform reveals temporal parameters including open quotient, skewing quotient, maximum flow declination rate, and peak flow amplitude that characterize laryngeal function.

Glottal source power, calculated from pressure-flow products or spectral power density, typically ranges from 1-10 milliwatts for conversational speech to 10-100 milliwatts for loud phonation, with only 0.1-1% ultimately radiated as acoustic power. The remaining energy dissipates through viscous losses in tissues (50-70%), absorption in vocal tract walls (10-20%), and other inefficiencies. Clinical applications include assessment of voice disorders (glottal insufficiency, hyperfunction, paralysis) and objective monitoring of treatment outcomes, while research applications enable parametric studies of laryngeal adjustments and validation of voice production models.

Despite challenges in vocal tract estimation and source-tract coupling assumptions, inverse filtering remains the primary non-invasive technique for measuring glottal source characteristics, providing essential insights into laryngeal function that radiated speech alone cannot reveal. Understanding glottal source power and its measurement through inverse filtering enables more precise characterization of voice disorders, more objective assessment of treatment effects, and deeper understanding of the biomechanical and aerodynamic basis of voice production.

Key Takeaways

✅ Inverse filtering separates glottal source from vocal tract filter by sequentially removing radiation and resonance effects
✅ Process involves integration (removing radiation) and inverse filtering (removing formants) to recover glottal flow waveform
✅ Recovered glottal flow reveals temporal parameters: open quotient, skewing quotient, MFDR, and peak flow amplitude
✅ Glottal source power ranges from 1-10 mW (conversational) to 10-100 mW (loud), calculated from pressure-flow products
✅ Only 0.1-1% of glottal source power ultimately radiates; most dissipates as heat in tissues and absorption
✅ Clinical applications include assessing glottal insufficiency, hyperfunction, and paralysis through objective waveform analysis
✅ Implementation challenges include accurate vocal tract estimation, source-tract coupling effects, and signal quality requirements
✅ Inverse filtering enables parametric studies of laryngeal function and validation of voice production models

Glottal Source Power and Inverse Filtering

Glottal Source Power and Inverse Filtering

The Source-Filter Framework

Fundamental Principle

Limitations of Direct Measurement

Inverse Filtering Technique

Basic Concept

Implementation Steps

Challenges and Sources of Error

Glottal Flow Waveform Characteristics

Temporal Parameters

Derived Parameters

Calculating Glottal Source Power

Time-Domain Power Calculation

Frequency-Domain Power Calculation

Typical Power Values

Relationship to Radiated Acoustic Power

Power Loss Mechanisms

Transfer Functions

Clinical Applications of Inverse Filtering

Voice Disorder Assessment

Treatment Monitoring

Research Applications

Parametric Studies

Voice Synthesis and Modeling

Summary

Key Takeaways

Related Topics

Further Reading