Automatic Detection of Laryngeal Pathologies in Records of Sustained Vowels by Means of Mel-Frequency Cepstral Coefficient Parameters and Differentiation of Patients by Sex

Abstract

Mel-frequency cepstral coefficients (MFCC) have traditionally been used in speaker identification applications. Their use has been extended to speech quality assessment for clinical applications during the last few years. While the significance of such parameters for such an application may not seem clear at first thought, previous research has demonstrated their robustness and statistical significance and, at the same time, their close relationship with glottal noise measurements. This paper includes a review of this parameterization scheme and it analyzes its performance for voice analysis when patients are differentiated by sex. While it is of common use for establishing normative values for traditional voice descriptors (e.g. pitch, jitter, formants), differentiation by sex had not been tested yet for cepstral analysis of voice with clinical purposes. This paper shows that the automatic detection of laryngeal pathology on voice records based on MFCC can significantly improve its performance by means of this prior differentiation by sex.

An insight to the automatic categorization of speakers according to sex and its application to the detection of voice pathologies: A comparative study

“…Results are given for male, female and mixed sexes, where the latter refers to combining male's and female's performance measures manually segmented sex-dependent system is 72.38%, mildly higher than the baseline sex-independent pathology detector which performed 71.65%. Results are in line with those in [27] where the classification accuracy of an automatic detector of pathology is lightly improved by using a manual segmentation of the database according to the sex of the speakers. Concerning the proposed sex-dependent pathology detector, and in comparison with the two previous baselines, a light performance improvement is also observed.…”

“…The number of Gaussians is varied in the following set: {4, 8, 16, 24, 32, 64, 128, 200, 256}. Training of the UBM model is carried out using a private database belonging to "Universidad Politécnica de Madrid" [27]. The performance of the classifiers is assessed using a 10-fold cross-validation strategy, calculating the classifier accuracy within a given confidence interval .…”

“…Table 1 comprises the final distribution of the data according to sex and condition after the speech therapist assessment. Similarly, the database employed for training the UBM models is presented in [27]. …”

“…Hence, the application of these systems for pathological voices is not demonstrated, being this a challenging problem due to the presence of perturbations inherent to pathological states. Respecting pathological voice detection, it is evidenced in [27] that a manual segmentation of the speakers' database according to their sex improves accuracy in an automatic pathology detection system. Furthermore, authors in [28] determined a significant sex-specific separation of control and pathological classes in the Saarbrücken database, by using statistical analysis and a series of acoustical and spectral measurements.…”

Computer Aided Recognition of Vocal Folds Disorders by Means of RASTA-PLP

“…(Marinus, Fechine, Gomes & Costa, 2009) and (Salhi, Talbi & Cherif, 2008) the authors have used artificial neural networks (ANN) to differentiate between different levels of pathology according to a perceptual quality voice scale. A study like (Fraile, Saenz-Lechon, Godino-Llorente, Osma-Ruiz & Fredouille, 2009) the patients were split out and differentiated by sex. The feature extraction used to train the ANN was based on MFCC yielding a classification accuracy of 88.3% with 53 normal and 173 pathological speakers from MEEI database.…”