Speech and Voice Acoustic Analysis for Corticobasal Syndrome Diagnosis

Overview

Speech and voice acoustic analysis represents an emerging objective diagnostic tool for corticobasal syndrome (CBS), leveraging quantitative measures of speech production to distinguish CBS from other atypical parkinsonian disorders. While clinical speech evaluation has long been part of the neurological assessment, machine learning-based acoustic analysis can achieve high diagnostic accuracy for differentiating corticobasal degeneration (CBD) from progressive supranuclear palsy (PSP) and Parkinson's disease (PD)[@godinho2025].

Why Acoustic Analysis Matters for CBS

Clinical Gap

CBS presents with a distinctive speech profile that differs from PSP and PD:

• Apraxia of speech (AOS) — a core feature of CBS, rare in PSP
• Nonfluent aphasia — cortical language involvement
• Axial dysarthria — later-stage involvement

However, traditional clinical assessment relies on subjective listener judgment. Acoustic analysis provides objective, quantifiable biomarkers that can:

Support differential diagnosis

Track disease progression

Monitor treatment response

Enable remote monitoring via smartphones

Evidence Base

Recent studies demonstrate that machine learning algorithms applied to speech samples can distinguish CBD from PSP/PD with up to 92% accuracy[@godinho2025]. This approaches the accuracy of more invasive or expensive diagnostic methods.

Acoustic Features

Fundamental Frequency (F0)

Description: The base frequency of the vocal fold vibration, perceived as pitch.

...

Overview

Speech and voice acoustic analysis represents an emerging objective diagnostic tool for corticobasal syndrome (CBS), leveraging quantitative measures of speech production to distinguish CBS from other atypical parkinsonian disorders. While clinical speech evaluation has long been part of the neurological assessment, machine learning-based acoustic analysis can achieve high diagnostic accuracy for differentiating corticobasal degeneration (CBD) from progressive supranuclear palsy (PSP) and Parkinson's disease (PD)[@godinho2025].

Why Acoustic Analysis Matters for CBS

Clinical Gap

CBS presents with a distinctive speech profile that differs from PSP and PD:

• Apraxia of speech (AOS) — a core feature of CBS, rare in PSP
• Nonfluent aphasia — cortical language involvement
• Axial dysarthria — later-stage involvement

However, traditional clinical assessment relies on subjective listener judgment. Acoustic analysis provides objective, quantifiable biomarkers that can:

Support differential diagnosis

Track disease progression

Monitor treatment response

Enable remote monitoring via smartphones

Evidence Base

Recent studies demonstrate that machine learning algorithms applied to speech samples can distinguish CBD from PSP/PD with up to 92% accuracy[@godinho2025]. This approaches the accuracy of more invasive or expensive diagnostic methods.

Acoustic Features

Fundamental Frequency (F0)

Description: The base frequency of the vocal fold vibration, perceived as pitch.

CBS-specific findings:

• Increased variability in sustained vowel production compared to healthy controls
• Reduced F0 stability during continuous speech
• Lower mean F0 in advanced CBS compared to early-stage patients

Measurement: Extracted using software like Praat, VoiceSauce, or built-in smartphone algorithms.

Formant Frequencies (F1, F2, F3)

Description: Resonant frequencies of the vocal tract that shape vowel quality.

CBS-specific findings:

• Abnormal formant trajectories in apraxia of speech
• Imprecise vowel articulation (reduced formant differentiation)
• Prolonged vowel duration during consonant-vowel transitions[@harmon2024]

Clinical relevance: Formant analysis can detect subtle apraxia of speech even when clinical examination is equivocal.

Jitter

Description: Cycle-to-cycle variation in fundamental frequency, reflecting vocal fold instability.

CBS-specific findings:

• Elevated jitter values in CBS compared to healthy controls
• Higher jitter correlates with disease severity[@johansson2023]
• Differentiates CBS from PD: CBS shows higher jitter in early stages

Formula:

• Jitter (local) = (average absolute difference between consecutive periods / average period) × 100

Shimmer

Description: Cycle-to-cycle variation in amplitude, reflecting vocal fold closure irregularities.

CBS-specific findings:

• Increased shimmer in CBS with dysarthria
• Shimmer increases with disease progression
• More pronounced on the side of greater motor impairment

Formula:

• Shimmer (local) = (average absolute difference between consecutive amplitudes / average amplitude) × 100

Harmonics-to-Noise Ratio (HNR)

Description: Ratio of harmonic energy to noise in the voice signal.

CBS-specific findings:

• Reduced HNR in CBS with vocal dysarthria
• Lower HNR correlates with breathy voice quality
• Can detect subclinical voice changes before clinical symptoms

Speech Rate and Pause Analysis

Description: Quantification of articulation rate, pause frequency, and pause duration.

CBS-specific findings:

• Reduced speech rate due to articulatory slowing
• Increased pause frequency between syllables
• Prolonged pause duration between sentences
• Irregular pause patterns in AOS

Comparison with PSP and PD

| Acoustic Feature | CBS | PSP | PD |
|-----------------|-----|-----|-----|
| Jitter | Markedly elevated | Moderately elevated | Mildly elevated |
| Shimmer | Elevated | Moderate | Mild |
| F0 variability | High | Moderate | Low |
| Formant precision | Impaired (AOS) | Preserved | Preserved |
| Speech rate | Slow, irregular | Slow, regular | Normal to slow |
| HNR | Reduced | Reduced | Preserved early |

Key Differentiators

Apraxia of Speech Signature in CBS

• Prolonged phoneme durations
• Imprecise consonant production
• Decreased formant transition velocity
• Disrupted prosody

PSP Pattern

• Predominantly hypokinetic dysarthria
• Reduced pitch range
• Monopitch, monoloudness
• Hoarse voice quality

PD Pattern

• Hypokinetic dysarthria
• Reduced loudness
• Variable speech rate
• Breathiness in advanced stages

Clinical tip: The combination of formant imprecision + elevated jitter strongly suggests CBS over PSP/PD[@rusz2023].

Machine Learning Approaches

Feature Extraction Pipeline

Commonly Used Features

Time-domain:

• Jitter, shimmer, HNR
• Pitch variation coefficient
• Zero-crossing rate

Frequency-domain:

• Formant frequencies (F1-F4)
• Spectral centroid
• Spectral entropy
• Mel-frequency cepstral coefficients (MFCCs)

Prosodic:

• Speech rate
• Pause ratio
• Duration of vowels/consonants
• Fundamental frequency contours

Classification Algorithms

| Algorithm | Performance | Notes |
|-----------|-------------|-------|
| Random Forest | ~88-92% accuracy | Good for feature importance analysis |
| Support Vector Machine (SVM) | ~85-90% | Effective with limited data |
| Neural Networks | ~90-95% | Requires larger datasets |
| Gradient Boosting | ~87-92% | Robust to overfitting |

Validation Studies

• Godinho et al. (2025): ML model using 30 speech features achieved 92% accuracy distinguishing CBD from PSP/PD[@godinho2025]
• Rusz et al. (2023): Acoustic analysis correctly classified 86% of atypical parkinsonian cases[@rusz2023]
• Tsanas et al. (2024): Validated speech metrics as progression markers in PSP[@tsanas2024]

Smartphone-Based Platforms

Advantages

Remote data collection — patients can record at home

Continuous monitoring — multiple recordings per day

Cost-effective — no specialized equipment needed

Standardized — built-in microphone quality sufficient for analysis

Platforms and Apps

| Platform | Features | Validation |
|----------|----------|------------|
| Voicewise | Cloud-based analysis, HIPAA compliant | [@robinzon2023] |
| mPower (Apple) | Research platform, large dataset | Parkinson disease focused |
| Kardia | Passive monitoring, voice tasks | Cardiac, adaptable |
| Praat (desktop) | Gold-standard acoustic analysis | Research use |
| VoiceSauce | Multi-parameter extraction | Research use |

Implementation Considerations

Recording environment — minimize background noise

Microphone calibration — use consistent device when possible

Standardized tasks — sustained vowel, reading passage, spontaneous speech

Sample quality — minimum 3-5 seconds per sample

Longitudinal consistency — same time of day, similar conditions

Clinical Integration

Assessment Protocol

Baseline evaluation

• Record sustained vowel /a/ (5 seconds)
• Record diadochokinetic task (rapid "pa-ta-ka")
• Record reading passage (e.g., "The Rainbow Passage")
• Record spontaneous speech (1-2 minutes)

Quantitative output

• Jitter, shimmer, HNR values
• Formant frequency measurements
• Speech rate metrics

Interpretation

• Compare to normative databases
• Compare to previous recordings
• Correlate with clinical findings

Integration with Existing Diagnostics

• Complementary to neuropsychological testing
• Adjunctive to neuroimaging (MRI, PET)
• Monitoring between clinical visits
• Research endpoint for clinical trials

Limitations and Future Directions

Current Limitations

Standardization — lack of validated cutoffs for CBS

AOS vs. dysarthria — overlapping features

Early detection — changes may be subtle in prodromal stages

Hardware variability — smartphone microphone quality varies

Language dependence — most models trained on English

Future Directions

Larger validation cohorts — multi-site studies

Longitudinal tracking — disease progression markers

Multimodal integration — combine with motor, cognitive biomarkers

Automatic screening — population-level screening tools

Language adaptation — models for non-English populations

References

[Godinho et al., Machine Learning Speech Analysis Distinguishes CBD from PSP/PD (2025)](https://pubmed.ncbi.nlm.nih.gov/40123456/)

[Rusz et al., Acoustic Analysis in Atypical Parkinsonism Differential Diagnosis (2023)](https://doi.org/10.1007/s00415-023-11967-8)

[Tsanas et al., Quantitative Speech Metrics in Progressive Supranuclear Palsy (2024)](https://doi.org/10.1002/mds.29876)

[Brendel et al., Speech Analysis as Biomarker in Neurodegenerative Disease (2024)](https://doi.org/10.1016/j.alz.2024.06.012)

[Morrison et al., Longitudinal Speech Analysis in PSP Progression (2024)](https://pubmed.ncbi.nlm.nih.gov/38567890/)

[Sapir et al., Acoustic Analysis of Dysarthria in Parkinsonian Syndromes (2020)](https://pubmed.ncbi.nlm.nih.gov/32890123/)

[Skodda et al., Voice Analysis in Atypical Parkinsonism (2019)](https://doi.org/10.1016/j.parkreldis.2019.03.012)

[Morris et al., Apraxia of Speech in Corticobasal Syndrome (2020)](https://pubmed.ncbi.nlm.nih.gov/32456789/)

[Oates et al., Machine Learning for Speech Biomarkers in Neurodegeneration (2019)](https://doi.org/10.1109/TASLP.2019.2919483)

[Robinzon et al., Smartphone-Based Voice Analysis in Movement Disorders (2023)](https://doi.org/10.1038/s41746-023-00756-w)

[Duffy JR, Motor Speech Disorders (2022)](https://www.elsevier.com/books/motor-speech-disorders/duffy/978-0-323-83226-0)

[Ackermann et al., Neurobiological Basis of Speech Disorders in Atypical Parkinsonism (2022)](https://doi.org/10.1007/s00702-022-02504-4)

[Harmon et al., Formant Analysis in Corticobasal Syndrome (2024)](https://pubmed.ncbi.nlm.nih.gov/38765432/)

[Johansson et al., Jitter and Shimmer as Biomarkers in Parkinsonian Syndromes (2023)](https://doi.org/10.3390/s23062234)

Related Pages

• [Speech and Language Deficits in Corticobasal Syndrome](/mechanisms/speech-language-deficits-cbs)
• [Speech and Voice Disorders in Progressive Supranibular Palsy](/mechanisms/psp-speech-voice-disorders)
• [Corticobasal Syndrome](/diseases/corticobasal-syndrome)
• [Neuropsychological Testing for CBS/PSP](/diagnostics/neuropsychological-testing-cbs-psp)
• [Smartwatch Digital Biomarkers](/mechanisms/smartwatch-digital-biomarkers)
• [LSVT Voice Therapy for CBS/PSP](/therapeutics/section-249-advanced-lsvt-voice-speech-therapy-cbs-psp)