Modulation Spectra Morphological Parameters: A New Method to Assess Voice Pathologies according to the GRBAS Scale

Area of Science:

• Speech Pathology
• Acoustic Analysis
• Signal Processing

Background:

• Perceptual evaluations of disordered voices by speech pathologists can be subjective and influenced by external factors, impacting assessment reliability.
• The development of automatic systems is crucial to enhance the objectivity and consistency of voice quality assessments.

Purpose of the Study:

• To present an automatic system for assessing the Grade and Roughness of speech based on the GRBAS perceptual scale.
• To evaluate the effectiveness of novel Modulation Spectra Morphological Parameters compared to classic Mel-Frequency Cepstral Coefficients (MFCCs) for voice disorder evaluation.

Main Methods:

• Utilized two parameterization methods: Mel-Frequency Cepstral Coefficients (MFCCs) and a new set of Modulation Spectra Morphological Parameters (MSMP).
• Employed Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) for feature dimensionality reduction.
• Applied Gaussian Mixture Model (GMM) classifiers to assess feature discrimination capabilities for different voice levels.

Main Results:

• The proposed Modulation Spectra Morphological Parameters achieved higher classification efficiencies: 81.6% for Grade and 84.7% for Roughness.
• Mel-Frequency Cepstral Coefficients (MFCCs) yielded efficiencies of 80.5% for Grade and 77.7% for Roughness.
• The MSMP demonstrated superior performance in distinguishing voice disorder levels compared to MFCCs.

Conclusions:

• The novel Modulation Spectra Morphological Parameters show significant promise for the automatic and reliable evaluation of Grade and Roughness in disordered speech.
• This automated approach can potentially improve the objectivity and consistency of voice assessments in clinical settings.