What is cepstral peak prominence (CPP) in voice analysis?

Cepstral peak prominence (CPP) is an acoustic measure of voice quality that quantifies the strength of the dominant peak in the cepstrum of a voice signal, relative to a regression line through the surrounding cepstrum. It reflects how clearly the harmonic structure of the voice stands out above the noise floor: higher CPP values indicate clearer voice quality, and lower values indicate dysphonia. It was introduced by Hillenbrand and colleagues in 1994 and identified in a 2009 meta-analysis as the most robust acoustic correlate of overall dysphonia severity.

What is the difference between CPP and CPPS?

CPPS (smoothed cepstral peak prominence) applies time and quefrency smoothing before the peak is measured, which reduces frame-to-frame variability. CPPS was introduced by Hillenbrand and Houde in 1996 and is the version implemented in Praat. The two are related but numerically different, and different programs implement them with different settings, so CPP and CPPS values from different software are not interchangeable.

What is a normal CPP value?

It depends entirely on the software and the task. For Praat, published cutoffs are 14.45 dB for sustained /a/ and 9.33 dB for read speech: values below these suggest a voice disorder (Murton et al., 2020). The same study's ADSV cutoffs are about 3 dB lower (11.46 and 6.11 dB), and an independent running-speech study using a third implementation found a cutoff of 4.0 dB (Heman-Ackah et al., 2014). Always compare a CPP value against a cutoff from the same software and the same task.

Why does ASHA recommend a cepstral measure over jitter and shimmer?

The 2018 ASHA expert panel selected a cepstral measure as the recommended acoustic correlate of voice quality because cepstral measures are viable across the entire range of dysphonia severity, in sustained vowels and in connected speech. Jitter and shimmer require period-by-period cycle identification, which fails in severely dysphonic voices — exactly the cases where objective measurement matters most — and are only valid for steady sustained vowels.

How is CPP measured in Praat?

Praat computes CPPS through a multi-step process: the voice signal is converted to a power cepstrum via Fourier analysis, smoothing is applied across time and quefrency, the dominant cepstral peak is identified within a defined quefrency range, a regression line is fit through the surrounding cepstrum, and CPPS is the dB difference between the peak and the regression line at that quefrency. The analysis settings affect the result, so settings should be reported. PhonaLab uses Praat's algorithm directly via Parselmouth, producing values identical to desktop Praat with the same settings.

Is CPP valid for telehealth and smartphone recordings?

For locally recorded, uploaded files: largely yes. CPP from smartphones and tablets with a headset microphone correlated at r of about .99 with a research-grade recording chain. For live videoconferencing audio: no. CPPS decreased on every platform tested, by 1.4 to 2.3 dB in connected speech, always toward the dysphonic side, so a live-capture CPPS compared against a published cutoff will over-call dysphonia. Record locally, upload the file, analyze the file.

Back to Guides

Understanding Cepstral Peak Prominence (CPP): The Anchor Measure of Modern Voice Assessment

November 15, 2025 (updated September 1, 2026)14 min readJorge C. Lucero

🎯 Key Takeaways

The 2018 ASHA instrumental protocol chose a cepstral measure as the acoustic correlate of voice quality, because it works across the entire severity range and in connected speech (Patel et al., 2018)
CPP works where perturbation measures fail: severely dysphonic voices and continuous speech, with no cycle-by-cycle pitch tracking needed
Published Praat cutoffs: below 14.45 dB (sustained /a/) or 9.33 dB (read speech) suggests a voice disorder, with up to 94.5% accuracy (Murton et al., 2020)
But cutoffs are software-specific: the same study's ADSV cutoffs are ~3 dB lower, and a third implementation puts its running-speech cutoff at 4.0 dB. A CPP value without a software name is not interpretable
Meta-analytic backing: smoothed CPP was identified as the most robust acoustic correlate of overall dysphonia severity (Maryn et al., 2009)
Telehealth caveat: live videoconferencing depresses CPPS by 1.4–2.3 dB toward the dysphonic side on every platform tested. Uploaded local recordings, yes; live platform audio against a cutoff, never

In recent years, cepstral peak prominence (CPP) has become the anchor acoustic measure for assessing voice quality. The 2018 ASHA expert panel on instrumental voice assessment selected a cepstral measure as the recommended acoustic correlate of voice quality—“a general measure of dysphonia”—in place of traditional perturbation measures like jitter and shimmer (Patel et al., 2018). The panel itself noted this was the most contentious choice in the entire protocol, and the recommendation remains current as of 2026. Yet many clinicians still reach first for the traditional measures—measures that fail precisely when we need them most: with severely dysphonic voices.

If you were trained on jitter and shimmer, you're not alone. Most speech-language pathology programs still emphasize them. But the field has shifted, and understanding CPP is now essential for evidence-based voice assessment.

Let me explain what CPP actually measures, where it came from, why the evidence favors it, and—most importantly—how to use it correctly in clinical practice: which cutoffs apply to which software, and what changes when the recording comes from a phone or a telehealth session.

The Problem with Traditional Measures

For decades, jitter (cycle-to-cycle pitch variability) and shimmer (cycle-to-cycle amplitude variability) were the go-to acoustic measures. They made intuitive sense: a smooth, periodic voice signal should have minimal perturbation, while a rough or hoarse voice would show increased variability.

But there's a fundamental problem: both measures require the software to identify individual glottal cycles.

The Catch-22 of Perturbation Measures

Jitter and shimmer work for mildly dysphonic voices, but fail for severely dysphonic voices—where you need objective measures most. When the voice becomes buried in noise (severe breathiness, roughness, or hoarseness), there isn't enough periodic information for the algorithm to identify individual vocal fold cycles, and the numbers it returns are artifacts.

Additional limitations of jitter and shimmer:

Valid only for steady sustained vowels: they can't assess voice quality during continuous speech, which is how patients actually communicate (Patel et al., 2018)
Inconsistent correlation with perception: meta-analytic correlations with roughness and breathiness vary widely across studies (Barsties v. Latoszek et al., 2018)
Software-dependent values: Praat and MDVP disagree by design, especially on noisy signals
Fragile under noise and loudness variation: ambient noise and vocal intensity both shift the values substantially

We cover these limits, the signal-typing rules, and the software problem in depth in our jitter and shimmer guide.

Enter Cepstral Peak Prominence (CPP)

CPP came out of signal processing research on breathiness. Hillenbrand, Cleveland, and Erickson introduced the measure in 1994; Hillenbrand and Houde extended it in 1996 with the smoothed variant (CPPS) and showed it worked on continuous speech, not just vowels. Clinical adoption followed (Heman-Ackah et al., 2002), and a 2009 meta-analysis of acoustic correlates of overall voice quality concluded that smoothed CPP was the most robust acoustic correlate of dysphonia severity across studies (Maryn et al., 2009). That meta-analysis is the evidence base the ASHA panel cited when it selected a cepstral measure in 2018.

What is CPP?

The cepstrum is, loosely, a spectrum of the log spectrum: it re-analyzes the spectrum to ask “how regularly spaced are the harmonics?” A voice with clear, evenly spaced harmonics produces a sharp peak in the cepstrum at the quefrency corresponding to the vocal fold vibration period. CPP measures how far that peak rises above a regression line through the rest of the cepstrum, in dB. No individual cycles need to be found—the measure looks at the signal as a whole, which is exactly why it keeps working when cycle identification breaks down.

Conceptual Framework

Think of the voice spectrum as a city skyline. In a healthy voice, harmonic peaks (the buildings) stand tall and distinct above the noise floor (the horizon). In dysphonia, increased aperiodic energy (fog) obscures these harmonics. CPP quantifies how prominently the harmonic structure rises above this noise floor.

Why CPP Works Better

Feature	Jitter/Shimmer	CPP
Cycle-by-cycle tracking	✗ Required (fails for severe dysphonia)	✓ Not required
Valid for continuous speech	✗ No (sustained vowels only)	✓ Yes (Hillenbrand & Houde, 1996)
Correlates with perceived severity	⚠ Varies widely across studies (Barsties v. Latoszek et al., 2018)	✓ Most robust correlate in meta-analysis (Maryn et al., 2009); r² up to .74 for estimating severity (Murton et al., 2020)
Valid across severity range	✗ Mild-to-moderate, Type 1 signals only	✓ Entire range (Patel et al., 2018)
Software-independent values	✗ No	✗ Also no—see the cutoffs below

Clinical Cutoffs (Software- and Task-Specific)

CPP vs. CPPS – and Why the Software Name Is Part of the Number

Praat computes CPPS (smoothed CPP), which averages across time and quefrency before measuring the peak (Hillenbrand & Houde, 1996; Maryn & Weenink, 2015). Other programs implement CPP or CPPS with different smoothing, windowing, and voicing-detection choices. The result: the same recording yields systematically different values in different software, and cutoffs do not transport. In Murton et al. (2020), the Praat cutoffs sit about 3 dB above the ADSV cutoffs on the same recordings.

Implementation & source	Sustained /a/	Connected speech
Praat / PhonaLab CPPS (Murton et al., 2020)	< 14.45 dB	< 9.33 dB (Rainbow Passage)
ADSV CPP (Murton et al., 2020)	< 11.46 dB	< 6.11 dB (Rainbow Passage)
CPPS, running speech (Heman-Ackah et al., 2014)	—	< 4.0 dB (sensitivity 92.4%, specificity 79%)

Values below the cutoff suggest a voice disorder; higher values indicate clearer voice quality. The Murton cutoffs classified disordered vs. healthy voices with up to 94.5% accuracy (295 patients, 50 controls). The Heman-Ackah study (835 patients, 50 controls) used a third implementation on running speech—its 4.0 dB cutoff is on a different scale entirely, which is the point: a CPP value means nothing without the software name and the task attached.

Both programs work—they just don't interchange

Sauder, Bretl, and Eadie (2017) compared CPPS from Praat and ADSV head to head for predicting voice disorder status: Praat reached an AUC of 0.91 (82% accuracy) and ADSV 0.81 (75% accuracy), with the two programs' values highly correlated (r = .88) yet numerically different. Either can support clinical screening. Mixing one program's values with the other's cutoffs cannot.

Two practical corollaries. First, report the task: vowel and connected-speech CPP sit on different scales (compare the columns above), and connected-speech analysis is also sensitive to how much unvoiced material the passage contains and how the program handles it. Second, report the settings: Praat's CPPS depends on its analysis parameters (Maryn & Weenink, 2015), so a defensible report names software, version, task, and settings—then compares only against cutoffs derived the same way.

CPP in the Real World: Phones, Telehealth, and Recording Quality

CPP is more robust to recording conditions than perturbation measures, but robust is not immune. Three situations deserve explicit rules:

Uploaded smartphone recordings: yes, with a headset

In the Bridge2AI-Voice validation, CPP from smartphones and tablets—with and without a low-cost headset microphone—correlated at r ≈ .99 with a research-grade chain (Awan et al., 2024). Offsets between devices remain, so pipeline-matched references still apply, but ranking and tracking are preserved. Protocol details in our smartphone recording guide.

Live telehealth audio: never against a cutoff

CPPS decreased on all six videoconferencing platforms tested, with large effect sizes—by 1.4–2.3 dB in connected speech, always downward, toward the dysphonic side of any published threshold (Weerathunge et al., 2021). A 2025 home-recording replication found the same: CPP from live Zoom capture failed validity criteria (Hwang et al., 2025). Live capture systematically over-calls dysphonia. Use the asynchronous workflow in our telehealth guide.

File requirements: lossless, adequate sample rate

Cepstral analysis needs the spectral bandwidth the algorithm expects. In PhonaLab, CPP requires an original sample rate of at least 10 kHz; files below that (common for messaging-app voice notes) return a flagged null rather than a plausible-looking number. Prefer WAV or FLAC at 44.1 or 48 kHz, recorded in a quiet room.

Common Questions About CPP

Q: Can I calculate CPP with free software?

Yes! Praat is free (though it has a learning curve). Our PhonaLab Voice Analyzer also provides free browser-based CPPS calculation with cited normative comparisons and PDF reports—no installation required, and audio is never stored.

Q: Why are my Praat/PhonaLab values higher than values I see in papers?

This is expected. Praat and PhonaLab calculate CPPS, and on the same recordings the Praat cutoffs sit about 3 dB above ADSV's (Murton et al., 2020). Papers using different software report different ranges; a running-speech study with a third implementation works on a scale where 4.0 dB is the cutoff (Heman-Ackah et al., 2014). Always check which software and task a paper used, and apply software-matched values only.

Q: Is CPP valid for telehealth recordings?

Split the question in two. Locally recorded files uploaded for analysis: yes—smartphone CPP correlates at r ≈ .99 with research-grade recording when a headset and a quiet room are used (Awan et al., 2024). Live videoconferencing audio: no—every platform tested depressed CPPS by 1.4–2.3 dB toward the dysphonic side (Weerathunge et al., 2021), so a live-capture value read against a normative cutoff will over-call dysphonia. Record locally, upload, analyze the file.

Q: Should I report CPP to referring physicians?

Absolutely. Many ENTs are familiar with CPP from the literature. Include a brief interpretation with the software named: “CPPS (Praat) of 8.2 dB on sustained /a/ is below the published cutoff of 14.45 dB for distinguishing disordered from healthy voices (Murton et al., 2020).” The software name is part of the number.

Bottom Line: Why CPP Matters for Your Practice

1CPP is the acoustic voice-quality measure of the 2018 ASHA protocol—a choice grounded in meta-analytic evidence and still current in 2026
2It works where jitter and shimmer fail: moderate-to-severe dysphonia and continuous speech
3The software name is part of the number: Praat, ADSV, and other implementations sit on different scales, and cutoffs do not transport
4Track trends over time with a constant recording chain, task, and software, rather than fixating on single values
5Use alongside perceptual assessment and laryngoscopy for comprehensive evaluation—CPP quantifies severity, it does not diagnose
6Mind the recording chain: uploaded lossless files yes; live videoconferencing audio against a cutoff, never

🎤 Calculate CPP on Your Patient Recordings

Upload any voice recording and get instant CPPS analysis computed with Praat's algorithm, compared against published software-matched cutoffs with the citation shown, plus professional PDF reports. Files that cannot support cepstral analysis return a flagged null with the reason. Audio is processed in memory and never stored.

Try Free Voice Analyzer →

Includes CPP/CPPS, F0, jitter, shimmer, HNR, and AVQI multiparametric assessment

⚠️ Clinical Documentation Tool

The information in this article is provided for educational purposes and clinical documentation support. Acoustic measures like CPP 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 based on the complete clinical picture. PhonaLab tools are designed for professional use and do not provide medical diagnoses.

References & Further Reading

Patel RR, Awan SN, Barkmeier-Kraemer J, Courey M, Deliyski D, Eadie T, Paul D, Švec JG, Hillman R. (2018). Recommended protocols for instrumental assessment of voice: American Speech-Language-Hearing Association expert panel to develop a protocol for instrumental assessment of vocal function. American Journal of Speech-Language Pathology, 27(3), 887–905. doi:10.1044/2018_AJSLP-17-0009
Hillenbrand J, Cleveland RA, Erickson RL. (1994). Acoustic correlates of breathy vocal quality. Journal of Speech and Hearing Research, 37(4), 769–778.
Hillenbrand J, Houde RA. (1996). Acoustic correlates of breathy vocal quality: Dysphonic voices and continuous speech. Journal of Speech and Hearing Research, 39(2), 311–321.
Maryn Y, Roy N, De Bodt M, Van Cauwenberge P, Corthals P. (2009). Acoustic measurement of overall voice quality: A meta-analysis. The Journal of the Acoustical Society of America, 126(5), 2619–2634.
Maryn Y, Corthals P, Van Cauwenberge P, Roy N, De Bodt M. (2010). Toward improved ecological validity in the acoustic measurement of overall voice quality: Combining continuous speech and sustained vowels. Journal of Voice, 24(5), 540–555.
Maryn Y, Weenink D. (2015). Objective dysphonia measures in the program Praat: Smoothed cepstral peak prominence and Acoustic Voice Quality Index. Journal of Voice, 29(1), 35–43. doi:10.1016/j.jvoice.2014.06.015
Murton O, Hillman R, Mehta D. (2020). Cepstral peak prominence values for clinical voice evaluation. American Journal of Speech-Language Pathology, 29(3), 1596–1607. doi:10.1044/2020_AJSLP-20-00001
Heman-Ackah YD, Michael DD, Goding GS Jr. (2002). The relationship between cepstral peak prominence and selected parameters of dysphonia. Journal of Voice, 16(1), 20–27. doi:10.1016/S0892-1997(02)00067-X
Heman-Ackah YD, Sataloff RT, Laureyns G, Lurie D, Michael DD, Heuer R, Rubin A, Eller R, Chandran S, Abaza M, Lyons K, Divi V, Lott J, Johnson J, Hillenbrand J. (2014). Quantifying the cepstral peak prominence, a measure of dysphonia. Journal of Voice, 28(6), 783–788. doi:10.1016/j.jvoice.2014.05.005
Sauder C, Bretl M, Eadie T. (2017). Predicting voice disorder status from smoothed measures of cepstral peak prominence using Praat and Analysis of Dysphonia in Speech and Voice (ADSV). Journal of Voice, 31(5), 557–566. doi:10.1016/j.jvoice.2017.01.006
Barsties v. Latoszek B, Maryn Y, Gerrits E, De Bodt M. (2018). A meta-analysis: Acoustic measurement of roughness and breathiness. Journal of Speech, Language, and Hearing Research, 61(2), 298–323.
Weerathunge HR, Segina RK, Tracy L, Stepp CE. (2021). Accuracy of acoustic measures of voice via telepractice videoconferencing platforms. Journal of Speech, Language, and Hearing Research, 64(7), 2586–2599. doi:10.1044/2021_JSLHR-20-00625
Awan SN, Bahr R, Watts S, Boyer M, Budinsky R, Bridge2AI Voice Consortium, Bensoussan Y. (2024). Validity of acoustic measures obtained using various recording methods including smartphones with and without headset microphones. Journal of Speech, Language, and Hearing Research, 67(6), 1712–1730. doi:10.1044/2024_JSLHR-23-00759
Hwang K, van Brenk F, McAuliffe MJ, Choi J, Švec JG, Chang YHM, Keller B, Levy ES. (2025). Validity of acoustic speech measures obtained through videoconferencing with children with dysarthria. International Journal of Speech-Language Pathology. Advance online publication. doi:10.1080/17549507.2025.2563845
ASHA Practice Portal: Voice Disorders – Assessment and Treatment

Back to Guides