A noninvasive estimation of hypernasality using a linear predictive model.

Area of Science:

• Speech-Language Pathology
• Acoustic Analysis
• Biomedical Engineering

Background:

• Hypernasality, characterized by abnormal speech resonance, often results from soft palate defects.
• Current clinical assessments for hypernasality are often invasive or require costly instrumentation.
• Accurate quantification of hypernasality is crucial for evaluating surgical outcomes.

Purpose of the Study:

• To introduce a novel, quantitative method for estimating hypernasality.
• To develop a cost-effective and less invasive alternative to existing assessment tools.
• To correlate the proposed method's results with established hypernasality measures.

Main Methods:

• Utilizing a microphone and personal computer with a sound card to capture digitized speech data.
• Leveraging the inaccuracy of linear predictive models in vocal tract systems with frequency response zeros, a hallmark of hypernasality.
• Comparing sequences of linear predictive cepstrum from low- and high-order models to quantify hypernasality.

Main Results:

• The proposed method demonstrated a stronger correlation (0.58) with nasometer-measured nasalance compared to the Teager method (0.44).
• For data with nasalance exceeding 35%, the proposed method achieved a higher correlation (0.84) than the Teager method (0.71).
• The technique relies solely on digitized speech, indicating reduced invasiveness.

Conclusions:

• The proposed method offers a simple, cost-effective, and less invasive means of evaluating hypernasality.
• This approach provides a valuable tool for speech-language pathologists and researchers.
• Further validation may lead to widespread clinical adoption for assessing speech resonance disorders.