Harmonics-to-Noise Ratio (HNR): What It Tells You (And When It Doesn't)
🎯 Key Takeaways
- HNR measures signal clarity—the ratio of periodic (harmonic) energy to aperiodic (noise) energy in the voice
- Normal range: ~15–20+ dB in Praat; values below 10 dB suggest clinically significant noise
- Best predictor of breathiness—accounts for 30–50% of variance in perceptual breathiness ratings
- Fails with severe dysphonia—requires quasi-periodic signal; unreliable for highly aperiodic voices
- CPP is now preferred for overall dysphonia assessment; use HNR alongside CPP, not instead of it
Harmonics-to-Noise Ratio (HNR) is one of the foundational acoustic measures in voice assessment—and one of the most misunderstood. The concept seems simple: measure how much of the voice signal is clean, periodic sound versus turbulent noise. Higher values mean a clearer voice; lower values suggest breathiness or roughness.
But if you've ever compared HNR values across different software packages, you've likely noticed they don't match. And if you've tried to interpret HNR for a severely dysphonic patient, you may have gotten results that seemed clinically meaningless.
This guidewill explain what HNR actually measures, when it provides clinically useful information, when it fails, and how it relates to CPP—the measure that has largely supplanted it as the primary acoustic index of voice quality.
What Question Does HNR Ask?
Every acoustic measure asks an implicit question about the voice signal. For HNR, that question is:
"How much of this signal is clean, periodic vibration versus turbulent noise?"
A healthy voice produces a waveform that repeats nearly identically from cycle to cycle—this is periodicity. The vocal folds vibrate regularly, producing energy concentrated at the fundamental frequency and its integer multiples (harmonics). This periodic energy is what we perceive as tonal voice quality.
Noise enters the signal when something disrupts this regularity: turbulent airflow through an incompletely closed glottis (breathiness), irregular vocal fold vibration (roughness), or external interference (recording environment). This aperiodic energy spreads across the frequency spectrum rather than concentrating at harmonic frequencies.
HNR quantifies the balance between these two components, expressed in decibels:
HNR = 10 × log₁₀(Harmonic Energy / Noise Energy)
If 99% of energy is harmonic and 1% is noise: HNR = 10 × log₁₀(99/1) ≈ 20 dB
An HNR of 0 dB means equal energy in harmonics and noise—a substantially degraded signal. Healthy voices typically show HNR values of 15–20 dB or higher, indicating that harmonic energy dominates the signal by a factor of 30:1 to 100:1 or more.
How Is HNR Calculated?
HNR estimation can be performed in the time domain (using autocorrelation) or the frequency domain (using spectral analysis). Praat, the most widely used software for voice analysis, offers both methods:
Cross-Correlation Method
Preferred in Praat. Measures how well the signal correlates with a time-shifted version of itself. High correlation at the period lag indicates strong periodicity (high HNR).
Autocorrelation Method
Based on Boersma's (1993) algorithm. Identifies the fundamental period from autocorrelation peaks, then estimates harmonic vs. noise energy.
The key principle is the same: a periodic signal correlates strongly with itself at lags corresponding to its fundamental period. The ratio of this peak correlation to the uncorrelated (noise) component gives us HNR.
Technical Note: Pitch Tracking Dependency
HNR calculation requires accurate fundamental frequency (F0) detection. If the pitch tracker fails—common in severely aperiodic voices—HNR estimation becomes unreliable or impossible. This is a fundamental limitation, not a software bug.
Normative Values: The Messy Reality
If you search for "normal HNR values," you'll find a confusing range of numbers. This isn't carelessness—it reflects real methodological differences that clinicians must understand.
| Population | Typical HNR Range | Notes |
|---|---|---|
| Healthy adults | 15–20+ dB | Praat method; sustained /a/ |
| Females | Slightly higher than males | Some studies report 1–2 dB difference |
| Children | ~9–12 dB | Lower than adults; developing larynx |
| Older adults (70+) | May decrease with age | Presbyphonia effects |
| Clinical concern | <10 dB | Suggests significant noise component |
| Pathological | <7 dB | Yumoto et al. threshold; some researchers use this |
The Software Problem
HNR values from different software packages are not directly comparable. Studies comparing MDVP, CSL, Praat, and other systems found poor inter-program correlations (r = 0.4–0.6). One study reported the same voice samples yielding HNR values of 8.58 dB, 14.34 dB, and 18.48 dB across different programs—all supposedly measuring the same thing.
Clinical implication: Always use the same software for longitudinal tracking. Don't compare HNR values across platforms.
Vowel matters too. HNR varies significantly by vowel. The vowel /u/ typically produces higher HNR values than /a/ or /i/ because its spectral energy is concentrated at lower frequencies, where noise has less impact. If you're comparing sessions, use the same vowel task.
What Does HNR Tell You Clinically?
HNR correlates most strongly with the perceptual quality of breathiness. This makes physiological sense: incomplete glottal closure allows turbulent airflow through the glottis, adding noise energy to the voice signal.
HNR and Perceptual Voice Quality (GRBAS Scale)
Research shows NHR (inverse of HNR) accounts for 30–50% of variance in perceptual breathiness ratings.
Combined interpretation: Low HNR with elevated jitter and shimmer often indicates roughness rather than pure breathiness—the noise comes from irregular vibration rather than air escape. Low HNR with relatively normal perturbation measures suggests breathiness from glottal insufficiency.
When HNR Is Clinically Useful
- Tracking treatment progress for conditions with breathiness component (vocal fold paralysis, presbyphonia, nodules)
- Pre/post surgical comparison (medialization thyroplasty, injection laryngoplasty)
- Mild to moderate dysphonia where the voice remains quasi-periodic
- Component of multiparametric indices (AVQI, where it's one of 6 parameters)
- Screening vocal aging—some studies suggest HNR is more sensitive to aging than jitter
When HNR Fails: Critical Limitations
HNR has fundamental limitations that clinicians must understand. These aren't software bugs—they're inherent to the measure itself.
When HNR Becomes Unreliable
1. Severely Dysphonic Voices
HNR requires quasi-periodic signals. When a voice is so disordered that pitch tracking fails, HNR calculation becomes mathematically impossible or produces spurious values. This is precisely when objective measurement matters most—and when HNR can't help.
2. Type 3 and Type 4 Signals
Using Titze's voice signal classification: Type 1 (nearly periodic) and Type 2 (subharmonics/modulations) can be analyzed with HNR. Type 3 (chaotic) and Type 4 (random/noise) signals violate the assumptions HNR requires.
3. Smartphone and Mobile Recordings
HNR is sensitive to recording quality. Studies show significant bias when comparing HNR from mobile devices versus studio microphones. F0 and jitter are more robust; HNR and shimmer are not.
4. Background Noise
Environmental noise below 30 dB signal-to-noise ratio significantly degrades HNR measurements. In noisy clinical settings or telehealth contexts, HNR may reflect room acoustics more than voice quality.
5. Connected Speech
HNR is designed for sustained vowel analysis. Unlike CPP (which works well on running speech), HNR from connected speech samples is more variable and harder to interpret.
The Paradox of Perturbation Measures
HNR (like jitter and shimmer) works best on mildly disordered voices that are close to normal. The more severe the pathology, the less reliable the measure becomes. This is why researchers have increasingly turned to cepstral measures like CPP, which remain valid across a wider range of signal types.
HNR vs. CPP: When to Use Which
Cepstral Peak Prominence (CPP) has emerged as the preferred acoustic measure for overall dysphonia assessment. But this doesn't mean HNR is obsolete—they answer slightly different questions and have different strengths.
| Feature | HNR | CPP/CPPS |
|---|---|---|
| What it measures | Harmonic vs. noise energy ratio | Strength of periodic component in cepstrum |
| Best for | Breathiness specifically | Overall dysphonia severity |
| Sustained vowels | ✓ Good | ✓ Good |
| Connected speech | ⚠️ Limited | ✓ Good (CPPS) |
| Severe dysphonia | ✗ Fails | ✓ Still valid |
| Recording quality sensitivity | High (sensitive to noise) | Lower (more robust) |
| Vocal tract effects | Less sensitive | Affected by nasalization, intensity |
| Software agreement | Poor across platforms | Better (still imperfect) |
Clinical Recommendation
Use CPPS as your primary measure of overall dysphonia severity. Add HNR as a secondary measure when breathiness is a specific clinical concern. This combined approach is exactly what AVQI does—it includes both CPPS and HNR as components because they capture complementary aspects of voice quality.
One interesting finding from research: CPP is sensitive to vocal tract configuration (nasalization affects it), whereas HNR is not. This means CPP may capture both source and filter characteristics, while HNR more purely reflects glottal source quality. For certain clinical questions—like isolating the effect of vocal fold surgery from resonance changes—this distinction may matter.
HNR in Multiparametric Indices
Rather than using HNR in isolation, modern voice assessment often incorporates it into composite indices that combine multiple parameters. This approach acknowledges that no single measure captures the full complexity of voice quality.
HNR in AVQI
The Acoustic Voice Quality Index (AVQI) includes HNR as one of its 6 component parameters, alongside CPPS, shimmer (local and dB), slope, and tilt. By combining these measures using regression weights derived from perceptual ratings, AVQI provides a more robust overall dysphonia score than any single parameter.
In this context, HNR doesn't need to stand alone—its contribution is weighted appropriately against other measures, reducing the impact of its limitations.
Similarly, machine learning models for automatic dysphonia assessment consistently identify CPPS, HNR, and AVQI among the most influential features. Recent research using gradient boosting algorithms achieved near-expert agreement with perceptual ratings, with HNR ranking among the top predictors alongside cepstral measures.
Practical Recommendations
1. Recording Protocol
Use sustained /a/ at comfortable pitch and loudness, minimum 3 seconds. Record in a quiet environment (<50 dB ambient noise) with consistent microphone distance (15–30 cm). For telehealth, acknowledge that HNR may be less reliable than F0 or CPP.
2. Software Consistency
Always use the same software for longitudinal tracking. Praat is free, widely used, and produces consistent results. Don't compare HNR values between Praat and MDVP—they won't match.
3. Interpretation Guidelines
Look for changes over time rather than absolute values. An increase of 3–5 dB in HNR typically represents clinically meaningful improvement. Values below 10 dB warrant attention; values below 7 dB are clearly abnormal.
4. Combine with Other Measures
Never interpret HNR in isolation. Use it alongside CPP/CPPS, perceptual assessment (CAPE-V or GRBAS), and patient self-report (VHI). Consider multiparametric indices like AVQI for comprehensive assessment.
5. Know When to Trust It
Trust HNR more for mild-moderate dysphonia with intact periodicity. Trust it less for severe dysphonia, mobile recordings, or noisy environments. If the pitch track looks unstable, the HNR value is likely unreliable.
Frequently Asked Questions
Q: Is HNR the same as SNR (Signal-to-Noise Ratio)?
No, but they're related concepts. HNR specifically measures harmonic (periodic) energy versus aperiodic energy in the voice signal itself. SNR refers to the signal level relative to background/environmental noise. A recording can have good SNR (quiet room) but low HNR (breathy voice).
Q: What about NHR (Noise-to-Harmonics Ratio)?
NHR is simply the inverse of HNR. Some software (notably MDVP) reports NHR instead of HNR. Higher NHR means more noise (worse quality), while higher HNR means more harmonics (better quality). Check which metric your software reports to interpret correctly.
Q: Why does my patient's HNR differ between vowels?
Spectral content matters. Vowels like /u/ concentrate energy at lower frequencies, producing higher HNR. Vowels like /a/ and /i/ have more high-frequency energy, where jitter effects are more pronounced, leading to lower HNR. For consistency, always use the same vowel.
Q: Can I use HNR for voice screening?
Yes, for quick screening of mild-moderate issues. But remember its limitations: it may miss severe dysphonia (where calculation fails) and is sensitive to recording conditions. For robust screening, CPPS or multiparametric indices like AVQI are more reliable.
Q: How much HNR change is clinically meaningful?
Changes of 3–5 dB are typically noticeable. Smaller changes may fall within measurement variability. For clinical significance, combine HNR changes with perceptual improvement and patient-reported outcomes—don't rely on the number alone.
Bottom Line: HNR's Place in Modern Voice Assessment
- 1HNR measures signal clarity—the balance between periodic harmonics and aperiodic noise
- 2Best single predictor of breathiness—use it when glottal insufficiency is the clinical question
- 3Requires quasi-periodic signal—fails for severe dysphonia where pitch tracking breaks down
- 4Sensitive to recording conditions—less robust than CPP for telehealth or mobile recordings
- 5Use alongside CPP, not instead of it—together they capture complementary aspects of voice quality
📊 Analyze HNR and More with PhonaLab
Our Voice Analyzer calculates HNR alongside CPP, jitter, shimmer, F0, and other key parameters. Get instant analysis with clinical interpretation and PDF reports—all validated against Praat algorithms.
Try Free 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 HNR 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
- 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.
- Ferrand CT. (2002). Harmonics-to-noise ratio: an index of vocal aging.Journal of Voice, 16(4), 480-487.
- Bielamowicz S, Kreiman J, Gerratt BR, Dauer MS, Berke GS. (1996). Comparison of voice analysis systems for perturbation measurement. Journal of Speech and Hearing Research, 39(1), 126-134.
- Madill C, Nguyen D, et al. (2019). The impact of nasalance on cepstral peak prominence and harmonics-to-noise ratio. The Laryngoscope, 129(8), E299-E304.
- Bottalico P, Codino J, et al. (2020). Reproducibility of voice parameters: the effect of room acoustics and microphones. Journal of Voice, 34(3), 320-334.
- Little MA, Costello DA, Harries ML. (2009). Objective dysphonia quantification in vocal fold paralysis: comparing nonlinear with classical measures. Nature Precedings, 1.
- Goy H, Fernandes DN, Pichora-Fuller MK, van Lieshout P. (2013). Normative voice data for younger and older adults. Journal of Voice, 27(5), 545-555.
Dr. Jorge C. Lucero
Professor of Computer Science, University of Brasília
Dr. Lucero has 30+ years researching voice production and vocal fold dynamics. PhonaLab's acoustic analysis is built on Praat algorithms (via Parselmouth), ensuring consistency with research-standard methods.