A comparison of high precision F0 extraction algorithms for sustained vowels

Area of Science:

• Speech science
• Clinical acoustics
• Voice pathology analysis

Background:

• Perturbation analysis of sustained vowel waveforms is vital for evaluating pathological voices and monitoring treatment progress.
• Accurate estimation of voice fundamental frequency (F0) is a prerequisite for reliable perturbation analysis.
• Existing F0 extraction algorithms must be robust to noise and frequency/amplitude modulations common in voice pathologies for clinical applicability.

Purpose of the Study:

• To evaluate the performance of seven distinct F0 extraction algorithms.
• To assess the robustness of these algorithms against background noise and signal modulations.
• To determine the suitability of F0 algorithms for clinical voice analysis.

Main Methods:

• Examined seven F0 algorithms: AMDF, AC, ACC, IFAC, CEP, HPS, and WM.
• Evaluated algorithms using sustained vowel samples from normal and pathological subjects.
• Investigated algorithm performance with synthetic vowel waveforms under varying noise and modulation conditions.

Main Results:

• Algorithm performance varied significantly in accuracy and robustness.
• Noise and modulations negatively impacted the performance of most algorithms.
• Specific algorithms demonstrated superior resilience to perturbations, indicating potential clinical utility.

Conclusions:

• No single F0 algorithm proved universally superior across all conditions.
• Algorithm selection should consider the specific characteristics of the voice data and potential perturbations.
• Further research is needed to optimize F0 extraction for complex pathological voice signals.