ABSTRACT

Aphasia affects an estimated 30 to 40 percent of stroke survivors, with particularly severe consequences in sub-Saharan Africa where specialist speech-language pathology services are scarce and case-fatality rates remain high (Akinyemi et al., 2021; Owolabi et al., 2023). This paper describes the design, implementation, and preliminary technical evaluation of SpeakAgain, a free, open-source web application that combines rule-based severity classification, large-language-model sentence completion, and Google text-to-speech synthesis to provide augmentative communication and structured home-based rehabilitation for persons with aphasia. The system uses an eight-item self-report assessment to classify users into six aphasia subtypes and a derived 0 to 5 severity scale, and delivers graded naming, reading, cloze, and sentence-building exercises calibrated to severity. A lightweight caregiver messaging layer built on the Brevo transactional email application programming interface pushes daily progress digests and crisis alerts. The technology stack relies entirely on free or freemium components suitable for low-bandwidth deployment in low- and middle-income settings. We present the system architecture, the classification logic, the exercise engine, a simulation of 12-week recovery trajectories based on published therapy-dose relationships, and a discussion of the ethical and engineering considerations for deployment in Nigeria and comparable settings. SpeakAgain is released under a permissive licence at github.com and is being piloted as the sixth of ten planned neurological rehabilitation tools.

Keywords: aphasia, stroke rehabilitation, augmentative and alternative communication, large language models, mHealth, digital health, Nigeria, low-resource settings