Towards an objective measure of speakers' intelligibility derived from the speech wave envelope

This study investigates the possibility of a relationship between amplitude modulation in the speech envelope and a speaker's intelligibility or articulatory clarity. It aims at developing an intelligibility measure called the Modulation Index (MI). Speech samples from several English speakers and one French speaker were recorded and digitized. Speakers were asked to produce speech under three articulatory conditions: Underarticulated, Normally Articulated, and Overarticulated. A computer program was developed for calculation of MI, based on the amount of amplitude modulation depth in the envelope of each digitized speech sample. The MI values so obtained were compared with the corresponding ratings from English-speaking listeners who judged the articulatory clarity of the recorded utterances. Results indicate that the relationship between the perceptual data and the Modulation Index in its present form is weak and non-monotonic. Several factors may have affected the results of the comparison between the MI values and the perceptual data. There are indications that speakers were not always successful in producing the intended articulatory conditions. Also, despite precautions, there were some differences in intensity and duration between utterances from the three conditions. It is concluded that there is some correlation between amplitude modulation in speech envelopes and speakers' intelligibility or articulatory clarity. However, the Modulation Index will require modification before it can become a useful tool. Some modifications were briefly explored, and possible further modifications to both the Modulation Index and the experimental design are suggested for future investigations.Medicine, Faculty ofAudiology and Speech Sciences, School ofGraduat