Phonetic relevance and phonemic grouping of speech in the automatic detection of Parkinson's Disease

Abstract

Literature documents the impact of Parkinson's Disease (PD) on speech but no study has analyzed in detail the importance of the distinct phonemic groups for the automatic identification of the disease. This study presents new approaches that are evaluated in three different corpora containing speakers suffering from PD with two main objectives: to investigate the influence of the different phonemic groups in the detection of PD and to propose more accurate detection schemes employing speech. The proposed methodology uses GMM-UBM classifiers combined with a technique introduced in this paper called phonemic grouping, that permits observation of the differences in accuracy depending on the manner of articulation. Cross-validation results reach accuracies between 85% and 94% with AUC ranging from 0.91 to 0.98, while cross-corpora trials yield accuracies between 75% and 82% with AUC between 0.84 and 0.95, depending on the corpus. This is the first work analyzing the generalization properties of the proposed approaches employing cross-corpora trials and reaching high accuracies. Among the different phonemic groups, results suggest that plosives, vowels and fricatives are the most relevant acoustic segments for the detection of PD with the proposed schemes. In addition, the use of text-dependent utterances leads to more consistent and accurate models.

Figure 1

Waveforms and spectrograms of a speaker with PD (newly diagnosed) and a control speaker pronouncing the word [petaka]. Obtained from the Neurovoz corpus. Red dot-lines mark the first four formants calculated with Praat software. (A) Idiopathic PD female speaker. Age: 59. UPDRS: 9. Span: 720 ms. (B) Control female speaker. Age: 59. Span: 857 ms

Figure 2

Waveforms and spectrograms of a speaker with PD (in an advanced stage) and a control speaker pronouncing the word [petaka]. Obtained from the Neurovoz corpus. Red dot-lines mark the first four formants calculated with Praat software. (A) Idiopathic PD female speaker. Age: 85. UPDRS: 47. Span: 810 ms. (B) Control female speaker. Age: 83. Span: 780 ms.

Figure 3

First proposed approach. Phonemic grouping methodology is applied to the parkinsonian corpus (raw-phon).

Figure 4

Second proposed approach. Phonemic grouping methodology is applied to the parkinsonian and UBM corpora (phon-phon).

Figure 5

Third proposed approach. Phonemic grouping methodology is applied to the UBM corpus (phon-raw).

Figure 6

Representation of Gaussians in the third approach (phon-raw). The GMM-UBM in the example contains 7 Gaussians modelling 2 features. Plosive grouping was applied to the UBM which was adapted with all the frames from a DDK task.

Figure 7

Diagram of trials. In the diagram, the classifiers are GMM-UBM where the UBM is trained with Albayzin. (A) Scheme of cross-validation trials (11 folds). The classifiers can be referred to any type of phonemic grouping (fricative, liquid, nasal, plosive or vowels) or approach (baseline, raw-phon, phon-phon or phon-raw). (B) Scheme of cross-corpora trials. The classifiers can be referred to any type of phonemic grouping. Only the baseline and the proposed approach leading to the best results in cross-validation trials were used in cross-corpora trials.

Figure 8

Best accuracies (A) and AUC (B). Results are referred to the three proposed approaches (marked as 1, 2 or 3) and speech tasks, where mon stands for monologues and cross for cross-corpora trials.