Fuzzy Classifier for Speech Assessment in Speech Rehabilitation

Статья в журнале

The article describes the use of a fuzzy classifier as a mechanism for combining the values calculated from three metrics to obtain a quantitative assessment of the intelligibility of the syllable pronunciation. The resulting assessment is used to assess speech in speech rehabilitation after the treatment of the oral cavity and oropharynx oncology. The evaluation is based on the calculation of the distance as the similarity degree between the syllable pronunciation in the estimated recording (the patient’s speech in the process of rehabilitation) and the syllable pronunciation in the reference recording (the patient’s speech before surgery). It is proposed to use a fuzzy classifier with the genetic optimization algorithm NSGA II. The training datasets are quantitative and expert estimates of a set of syllable pronunciation recordings provided by the Cancer Research Institute Tomsk NRMC. Three datasets were used: 1020 recordings for each of the problematic phonemes [k], [s], [t]. Separate fuzzy classifiers were trained for each of the datasets. Classification accuracies are shown on the initial datasets and rebalanced datasets. The NeighborhoodCleaningRule and SMOTEENN algorithms were used for rebalancing the data. It was concluded that it is possible to use a fuzzy classifier as a combination mechanism to obtain the intelligibility assessment of the syllable pronunciation.

Журнал:

• Lecture Notes in Computer Science
• Springer (Томск)
• Индексируется в Scopus

Библиографическая запись: Fuzzy Classifier for Speech Assessment in Speech Rehabilitation [Electronic resource] / D. Novokhrestova [et al.] // Lecture Notes in Computer Science.–2022. – Vol. 13721. – P. 522-532. – DOI: 10.1007/978-3-031-20980-2_44