A Preliminary Study of Text-to-Speech Synthesizers Used by Dyslexics

Today a paper on assistive technology solutions for people suffering from dyslexia was presented at the 10th International Conference on Computers Helping People with Special Needs (ICCHP), held on July 12-14 at the University of Linz, Austria, by the staff of University of Manchester, UK.

The paper provides an in-depth study of text-to-speech synthesizers used by dyslexics, on the basis of the soundness of its pronunciations. There are many solutions in the assistive technology segment for text-to-speech synthesizers, targeting the visually challenged, people with reading difficulties, people with learning disabilities and also those learning a new language. However, this paper’s focus is on the efficacy of a text-to-speech synthesizer designed especially for dyslexics. The intention here is to investigate how dyslexics use the synthesizers and if there are any discrepancies.

Speech synthesizers have evolved over the years, in terms of their degree of naturalness and understandability of the voice. However, these criteria do not assure clarity and consistency in the comprehension of the speech. Different dyslexics approach the text-to-speech system in different ways, depending on his/her need. To harness the system optimally, the information presented must be clear and free from errors. The user might get distracted from comprehending the actual meaning of the text, if he/she gets stuck with an incompetent system. This is where the underlying contention of the study lies, if text-to-speech synthesizers speak correctly.

Certain interventions were conducted for the study. First, four commercial text-to-speech synthesizers most commonly used by dyslexic people were selected. Two non-dyslexic evaluators were made to listen to 704 common English words and a set of 44 homographs on these systems. The evaluators came up with three lists of results for the words, the words that were pronounced correctly, incorrectly and partially correctly. It is the last list that might mislead a user with dyslexia while spell-checking or distract him while reading a text block.

It was seen that the evaluators differed on the overall level of correctness, however, their overall ranking of text-to-speech synthesizers was found to be more or less consistent. Also, the agreement between the evaluators when it came to partially correct or incorrect words was relatively often. The final results indicated that the degree of correct pronunciation of common isolated words was quite high by a range of text-to-speech synthesizers.

In the other part of the intervention, 44 homographs were selected randomly and different sentences that contrasted various pronunciations of homographs were formed. The evaluators listened to these sentences on all of the synthesizers, and classified them into correct, incorrect and partially correct lists. The results indicated some differences between the two evaluators, however the overall ranking of the synthesizers was more or less consistent. It was seen that the synthesizers were not very effective at resolving the homographs when words were not given in the valid context.

The results of this study have come as a surprise, given the preconception that there would be a lot of errors in the pronunciation aspect of the synthesizers. However, it doesn’t enable dyslexics to use text-to-speech systems perfectly, as a lot of it depends on how the user perceives and uses the information. It was clear that the system is efficient at resolving homophones when supplied with content, however, the spell-checking feature still needs improvement. Nonetheless, the study has laid the foundation for further research in this area and provides accessibility options to dyslexics.