Automatic evaluation of voice and speech intelligibility after treatment of head and neck cancer

Cancer of the head and neck and its medical treatment and management, can have a negative impact on how a person sounds and talks. For the speech pathologist, rating a person’s speech intelligibility and voice quality is an important part of patient management. Rating someone’s speech or voice, however, can be difficult task to perform objectively as a listener’s ratings are often inconsistent. Computerized ratings, on the other hand, are consistent.
This thesis has focused on developing automatic prediction models for speech intelligibility and voice quality assessment for two groups of speakers treated for head and neck cancer. The first group discussed in Part I of this thesis are people with advanced tumours in the head and neck. These people received a type of non-surgical cancer treatment, called concurrent chemoradiotherapy (CCRT). This type of treatment can affect a person’s voice and speech. The second group of people were treated for advanced tumors in the larynx. These people underwent a total laryngectomy (TL), in which the larynx (also known as the ’voice box’) is removed. After a TL, speaking is possible with the aid of a valve that redirects air past vibrating structures in the neck towards the mouth. This type of speech is called tracheoesophageal speech and it sounds very different to how a person sounded before the surgery.