Acoustic investigation of speech pathologies based on the discriminative paraconsistent machine (DPM). (January 2020)

• Record Type: Journal Article
• Authors: Fonseca, Everthon Silva; Guido, Rodrigo Capobianco; Junior, Sylvio Barbon; Dezani, Henrique; Gati, Rodrigo Rosseto; Mosconi Pereira, Denis César
• Is Part Of: Biomedical signal processing and control. Volume 55(2020)
• Publication Date: 2020-01
• Subjects: Co-existent voice disorders -- Overlapped inter-class features -- Discriminative paraconsistent machine (DPM) -- Signal energy (SE) -- Zero-crossing rate (ZCR) -- Signal entropy (SH); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101615
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Innovatively, speech pathology detection is carried out based on paraconsistency. Discriminative Paraconsistent Machine (DPM) is the essence of the proposed approach. Feature extractions is based on energy, zero-crossing rate and entropy, providing relevant information from raw speech data. DPM allows for a wider inspection in contrast to binary results from usual classifiers. Abstract: Background: Voice disorders are related to both modest and severe health problems, including discomfort, pain, difficulty speaking, dysphagia and also cancer. Widely adopted worldwide, the combined invasive and subjective diagnosis of voice disorders is troublesome and error-prone. Contrarily, acoustic-based digital assessment allows for a non-intrusive and objective examination, stimulating the applications of computer-based tools. Objective: Consequently, this work describes a novel algorithm to investigate speech pathologies from the sounds of sustained vowels, particularly exploring a potential gap: the classification of co-existent issues for which the major phonic symptom is the same, implying in similar inter-class features. Method: By using the concepts of signal energy (SE), zero-crossing rates (ZCRs) and signal entropy (SH), which provide a joint time-frequency-information map, the proposed approach classifies voice signals based on the discriminative paraconsistent machine (DPM), allowing for the application of paraconsistency to treat indefinitions and contradictions. Results: An accuracy level of 95% was obtained under a subset of voices from the Saarbrucken voice database (SVD), with just a modest training. In complement, the proposed approach offers wider possibilities in contrast to current state-of-the-art systems, allowing for the inputs to be mapped into the paraconsistent plane in such a way that intermediary states can be found. Conclusion: Different from current algorithms, our technique focuses on a particular problem in the field of speech pathology detection (SPD), not yet explored in detail, proposing a way to successfully solve it. Furthermore, the results we obtained stimulate broaden studies involving speech data inconsistencies whilst providing a valid contribution.