Glottal-to-Noise Excitation (GNE): The Breathiness Measure That Isn't HNR
🎯 Key Takeaways
- GNE measures the source of noise, not just its amount—it asks whether the energy across frequency bands is driven by one common glottal pulse or by independent turbulence
- It isolates breathiness—GNE specifically detects turbulent (aspiration) noise, while HNR conflates turbulence with cycle-to-cycle irregularity
- It's a ratio from 0 to 1, not decibels—higher is healthier; values below ~0.87 fall outside the typical range in PhonaLab's PVQD-derived reference
- It needs a high sampling rate—GNE requires at least 10 kHz; below that it can return plausible-looking but invalid values without warning
- Use it alongside HNR and CPP—GNE answers a question neither of them does, which is why it survives selection across every perceptual dimension
Ask a Brazilian speech-language pathologist which acoustic measure they reach for when a voice sounds breathy, and many will name GNE before HNR. That preference isn't arbitrary. The glottal-to-noise excitation ratio was designed to answer a more specific question than HNR can—and for the clinical quality of breathiness, that specificity matters.
GNE is also one of the more misunderstood measures in the toolkit. It looks like a noise measure, sits next to HNR in most software outputs, and moves in the same direction (lower with more pathology). So clinicians reasonably assume it's just "another HNR." It isn't. The two measures can disagree, and when they do, the disagreement is clinically informative rather than a sign that one of them is wrong.
This guide explains what GNE actually measures, why it isolates turbulent noise where HNR cannot, how to read its values, the recording requirement that quietly invalidates it if you ignore it, and where it fits alongside the measures you already use. It's a companion to our HNR guide—reading both makes the distinction between them concrete.
What Question Does GNE Ask?
Every acoustic measure asks an implicit question about the signal. HNR asks how much of the signal is periodic. GNE asks something subtler:
"Is the energy in different frequency bands driven by the same underlying event—or by independent noise?"
Here is the physiological intuition. When the vocal folds close and open cleanly, each glottal pulse is a single, sharp excitation. That one event injects energy across the whole spectrum at the same instant. Because every frequency band is excited by the same pulse, the bands rise and fall together—their envelopes are correlated in time.
Turbulent airflow behaves differently. Air escaping through an incompletely closed glottis generates noise that is essentially random, and that randomness is independent from one frequency band to the next. The bands no longer move together. GNE quantifies exactly this: it measures how much the frequency bands share a common excitation. High shared excitation (clean glottal pulses) gives a high GNE; independent band activity (turbulence) drives it down.
Why this is different from "how much noise"
HNR can be lowered by anything that disrupts periodicity—turbulent air leakage or irregular, asymmetric vocal-fold vibration. GNE is largely indifferent to irregular vibration, because even irregular pulses still excite all bands together. What GNE responds to is the presence of a separate, incoherent noise source. That is what makes it a more targeted index of breathiness.
How Is GNE Calculated?
GNE was introduced by Michaelis, Gramss, and Strube in 1997. The computation, simplified, proceeds in four steps:
- 1The signal is inverse-filtered to remove the vocal-tract resonances, leaving an estimate of the glottal excitation.
- 2That excitation signal is split into a set of overlapping frequency bands using band-pass filters of fixed bandwidth.
- 3The Hilbert envelope of each band is computed, and cross-correlations are taken between band pairs whose centre frequencies are far enough apart to be independent under turbulence.
- 4GNE is the maximum of those correlation peaks: a single number between 0 and 1.
In Praat (and therefore in PhonaLab, via the Parselmouth bridge), GNE is computed with a fixed band configuration and the maximum of the resulting harmonicity object is extracted, consistent with the original definition. Holding that configuration constant is what makes the values comparable across recordings and tied to a meaningful reference range.
Technical Note: GNE Is Dimensionless
Unlike HNR, which is reported in decibels, GNE is a correlation-based ratio bounded between 0 and 1. A value of 1.0 would mean perfectly coherent band excitation (no turbulent noise); a value near 0 would mean fully incoherent excitation. Do not interpret GNE on a dB scale, and do not expect the same numeric thresholds that apply to HNR.
GNE vs. HNR: The Distinction That Matters
This is the heart of the measure. Both GNE and HNR drop when a voice becomes noisier, so on healthy and purely breathy voices they often agree. They part ways precisely on the cases where the distinction is most useful clinically.
HNR asks: how periodic?
Built on autocorrelation in the time domain. It falls whenever periodicity breaks down—from turbulent air leakage or from irregular, cycle-to-cycle vibration. It cannot tell you which.
GNE asks: one source or many?
Built on cross-band envelope correlation. It falls specifically when an independent noise source (turbulence) is present. Irregular but coherent pulses leave it largely intact.
The practical consequence is a two-by-two you can reason about at the bedside:
| Pattern | Likely mechanism | Perceptual correlate |
|---|---|---|
| GNE low, HNR low | Turbulent air leakage (glottal insufficiency) | Breathiness |
| GNE near-normal, HNR low | Irregular/asymmetric vibration without much leakage | Roughness more than breathiness |
| GNE low, HNR near-normal | Localized high-frequency turbulence; treat cautiously and re-check recording quality | Mild aspiration component |
| GNE high, HNR high | Coherent, low-noise phonation | Typical voice |
The clinically useful case
The second row is where GNE earns its place. A rough voice from irregular vibration can pull HNR down while leaving GNE comparatively preserved—telling you the noise is coming from vibratory irregularity, not air escape. That is a different therapeutic target. Neither measure alone tells you this; the contrast between them does.
Normative Values: Read Them in Context
Published GNE thresholds vary more than HNR thresholds do, and for a specific reason: GNE depends on the band configuration (lower/upper frequency, bandwidth) used to compute it. Two papers reporting "GNE" with different band settings are not reporting the same number. A normative cutoff is only meaningful when it is tied to a fixed computation.
PhonaLab's reference floor is derived empirically rather than imported from a paper with different settings. Using the healthy subset of the Perceptual Voice Qualities Database (PVQD)—speakers with a CAPE-V overall severity below 10—the 10th percentile of GNE converges to roughly 0.874–0.880 regardless of how the "healthy" boundary is drawn. PhonaLab uses 0.87 as the lower edge of the typical range for its specific band configuration.
| Reading | Typical GNE | Notes |
|---|---|---|
| Healthy adults | ≥ ~0.87 | PVQD-derived floor; PhonaLab band settings (500/4500/1000/80) |
| Sex difference | Negligible | Small effect (d ≈ 0.37, n.s.); a single threshold is defensible |
| Below typical | < 0.87 | Suggests a turbulent (breathy) noise component |
| Cross-study comparison | Not portable | Only compare GNE values computed with identical band settings |
The takeaway is not the digit 0.87 in isolation—it's the discipline behind it. Hold the computation fixed, derive the reference from a known population, and the threshold means something. Borrow a cutoff from a paper with different settings and it doesn't.
What GNE Tells You Clinically
GNE correlates most strongly with the perceptual quality of breathiness, which follows directly from what it measures: incomplete glottal closure produces the very turbulent, incoherent noise that drives GNE down.
Its reach is broader than that, though. In a PhonaLab analysis of the PVQD database (manuscript under review), GNE was selected as an informative parameter for every major perceptual dimension—it ranked first or second for breathiness, and it also entered the optimal parameter subsets for overall severity, roughness, and strain. Few measures survive selection across all four. That breadth is part of why it complements, rather than duplicates, HNR and CPP.
GNE Across Perceptual Dimensions (PVQD analysis)
Selected via orthogonal matching pursuit on 14 acoustic parameters; rankings reflect incremental predictive value, not raw correlation.
Combined interpretation: Pair GNE with HNR and the perturbation measures. Low GNE with low HNR and relatively normal jitter/shimmer points toward breathiness from glottal insufficiency. Low HNR with preserved GNE and elevated shimmer points toward roughness from irregular vibration. The measures are most powerful as a panel, not in isolation.
When GNE Is Clinically Useful
- Quantifying breathiness in glottal insufficiency, vocal fold paralysis, paresis, and presbyphonia
- Separating breathy from rough when HNR is reduced but the source of the noise is ambiguous
- Tracking change after medialization (injection or thyroplasty), where reducing air escape should raise GNE
- Component of breathiness indices—GNE is one of the parameters in the Acoustic Breathiness Index (ABI)
- Cross-checking HNR—agreement strengthens confidence; disagreement is diagnostically informative
When GNE Fails: Critical Limitations
GNE has limitations that, unlike HNR's, are not always obvious from the output. The first one is the most dangerous, because it fails silently.
The Sample-Rate Trap (the one most likely to mislead you)
GNE's upper analysis band sits in the mid-kHz range. For the signal to actually contain that band, its Nyquist frequency—half the sampling rate—must sit above it. That means GNE needs a sampling rate of at least about 10 kHz, and realistically the 16 kHz or higher you'd use for any spectral work.
Here is the trap: when given an under-sampled file, the underlying Praat command does not raise an error. It returns a plausible-looking number. An 8 kHz recording, whose Nyquist frequency falls below GNE's upper analysis band, can yield a value around 0.83 that looks like a real, mildly reduced GNE—when in fact the measurement is invalid. There is no red flag in the raw output.
This is a class of problem, not a one-off: any spectral-band measure asked to operate above the signal's Nyquist frequency can fail this way (CPP behaves similarly). PhonaLab guards against it by refusing to compute GNE below the required sampling rate rather than reporting a number it can't stand behind. If your software doesn't, check the sampling rate of every file before trusting a GNE value.
Other Situations Where GNE Is Unreliable
1. Band-setting dependence
Change the lower frequency, upper frequency, or bandwidth and the numeric value changes. GNE is only comparable across recordings analyzed with identical settings—and never directly comparable to a value from a paper that used different ones.
2. Severe, fully aperiodic voices
Like the perturbation measures, GNE assumes there is a glottal excitation to characterize. In Type 3 (chaotic) and Type 4 (noise) signals, that assumption breaks down and the value loses interpretability.
3. Not a standalone severity index
GNE is specific to the turbulent-noise component. A voice can be severely disordered along a roughness or strain dimension while GNE stays near normal. Don't read a single GNE value as overall severity.
4. Implementation and version sensitivity
Because GNE depends on filtering and envelope steps, different implementations (and even different software versions) can produce slightly different values. Reproducibility requires fixing the engine and its version—PhonaLab computes GNE through Praat 6.1.38 for exactly this reason.
A value is not a measurement
The deepest lesson GNE teaches is that software returning a number is not the same as the number being valid. A measure is only as trustworthy as the recording and the configuration behind it. Build the sampling-rate and settings check into your protocol, and GNE becomes one of your most informative tools. Skip it, and the tool will mislead you politely.
Bottom Line: GNE's Place in Voice Assessment
- 1GNE measures noise source, not just noise amount—shared glottal excitation versus independent turbulence
- 2It targets breathiness—where HNR conflates turbulence with vibratory irregularity, GNE isolates the turbulent component
- 3Read it as a 0–1 ratio in context—≥ ~0.87 is typical for PhonaLab's settings; thresholds aren't portable across band configurations
- 4Check the sampling rate first—below ~10 kHz GNE can return invalid values without any warning
- 5Use it as part of a panel—its agreement and disagreement with HNR and CPP is where the clinical value lives
📊 Analyze GNE and More with PhonaLab
Our Voice Analyzer reports GNE alongside HNR, CPP, jitter, shimmer, F0, and other key parameters—with a sampling-rate guard that refuses to report invalid values, and clinical interpretation in the PDF report. All measures are computed through Praat algorithms for research-grade consistency.
Open Voice Analyzer →No account required for Quick Voice Check • Full analysis with free account
⚠️ Clinical Documentation Tool
The information in this article is provided for educational purposes and clinical documentation support. Acoustic measures like GNE are intended to supplement—not replace—comprehensive voice evaluation including perceptual assessment, patient history, and laryngoscopic examination when indicated. All clinical decisions should be made by qualified healthcare professionals. PhonaLab tools do not provide medical diagnoses.
References & Further Reading
- Michaelis D, Gramss T, Strube HW. (1997). Glottal-to-noise excitation ratio—a new measure for describing pathological voices. Acta Acustica united with Acustica, 83(4), 700-706.
- Michaelis D, Fröhlich M, Strube HW. (1998). Selection and combination of acoustic features for the description of pathologic voices. The Journal of the Acoustical Society of America, 103(3), 1628-1639.
- Barsties von Latoszek B, Maryn Y, Gerrits E, De Bodt M. (2017). The Acoustic Breathiness Index (ABI): A multivariate acoustic model for breathiness. Journal of Voice, 31(4), 511.e11.
- Walden PR. (2022). Perceptual Voice Qualities Database (PVQD): Database characteristics.Journal of Voice, 36(6), 875.e15.
- Lucero JC. (2026). Algorithm verification and concurrent validity of a web-based platform for multiparametric acoustic voice quality indices. Journal of Voice. https://doi.org/10.1016/j.jvoice.2026.04.009
- Boersma P. (1993). Accurate short-term analysis of the fundamental frequency and the harmonics-to-noise ratio of a sampled sound. Proceedings of the Institute of Phonetic Sciences 17, 97-110.