Automatic speech recognition in voice-speech rehabilitation effectiveness evaluation in patients after laryngectomy

Introduction. Lost voice function compensation determines the personal and social life of laryngectomees. Automatic speech recognition and synthesis methods are widely used as apps for additional and alternative communication. One of the urgent tasks in clinical practice is voice restoration effectiveness evaluation. Aim. To evaluate the effectiveness of voice rehabilitation results in laryngectomized patients using the automatic assessment of speech intelligibility. Materials and methods. 3 groups of 30 laryngectomized patients depending on the method of voice rehabilitation (esophageal voice, tracheoesophageal voice, electrolarynx), and 14 patients after various surgical interventions on the larynx were included. All patients underwent pseudo-voice recording for further assessment of phrasal intelligibility using the automatic assessment software module, as well as by a trained and untrained listener. Results and discussion. Comparative analysis shows a minimum intelligibility level in automatic recognition, apparently due to semantic and contextual recognition even in untrained listener. Alaryngeal speech demonstrates worse recognition, compared to organ-preserving operations on the larynx, in untrained listeners and in automatic recognition. Experienced listeners demonstrate a consistently high level of recognition of all types of substitute speech. Conclusion. Objectification of the intelligibility assessment of substitute speech using automatic recognition systems allows leveling semantic and contextual recognition during assessment by both trained and untrained listeners. Automatic speech recognition and synthesys systems have application prospects in rehabilitation medicine, in particular, in patients with head and neck cancer.