Grandchild, speak clearly! Investigating algorithmic benefit for lisped speech understanding

Elderly listeners often have trouble understanding mispronounced speech. Specifically, this may happen in communication with grandchildren, who may have a lisp or simply talk quickly and unclearly. Lisping is a popular term for different types of consonant mispronunciation and is a common stage in speech development. These communication problems are worse in noise, since the content of the noise can mask the already ambiguously pronounced consonants. 

Here, we are interested in investigating the benefit of an adapted onset enhancement algorithm, as a method for improving speech understanding of lisped speech in noise. The algorithm was adapted to specifically enhance lisped speech onsets.

In order to investigate the effect of the algorithm, we recorded professional speech therapists from a speech therapy school in Nürnberg. The ability to produce different types of lisping is one of the speech therapist’s learned assets. We asked them to read the fairy tale "Nordwind und Sonne" and minimal word pairs, in normal pronunciation and in three different types of lisping. From these recordings, we developed a study protocol, consisting of two specific tests.

Seven normal hearing listeners participated in the study. Their subjective impression was assessed while listening to the fairy tale recordings. The objective performance was evaluated with the minimal word pairs test. In both tests the material was processed with and without the mentioned algorithm. The results of the study will be presented and discussed.