Acoustic-phonetic decoding for speech intelligibility evaluation in the context of Head and Neck Cancers

In addition to health problems, Head and Neck Cancers (HNC) can cause serious speech disorders that can lead to partial or complete loss of speech intel-ligibility in some patients. The clinician's evaluation of the intelligibility level before or after surgical treatment and / or during the rehabilitation phase is an important part of the clinical assessment. Perceptive assessment is the most widely used method in clinical practice to assess the level of intelligibility of a patient despite the limitations associated with it such as subjectivity and moderate reproducibility. In this paper, we propose to overcome these limitations by associating a specific task of speech production based on pseudo-words with an automatic speech processing system, both oriented towards acoustic-phonetic decoding. Compared to human perception, the automatic system reaches very high correlation rates and promising results when applied to a French speech corpus including 41 healthy speakers and 85 patients suffering from HNC.