## TL;DR
AI is democratizing accessibility — from real-time sign language translation on smartphones to AI-powered screen readers that describe visual scenes. Computer vision, speech recognition, and natural language processing collectively transform assistive technologies from specialized hardware into software running on the devices people already own.

## Core Explanation
AI accessibility applications: (1) Vision assistance — smartphone apps (Seeing AI, Lookout) use object detection and scene description to narrate the visual world for blind users; OCR reads text from signs, menus, and documents in real-time; (2) Hearing assistance — live transcription converts speech to text (Google Live Transcribe, Otter.ai); sign language recognition translates ASL gestures to text/speech; audio enhancement isolates speech from background noise using neural beamforming; (3) Mobility — AI-powered smart prosthetics learn user gait patterns and adapt in real-time; wheelchair path planning via computer vision obstacle detection; (4) Cognitive — personalized learning tools for dyslexia, ADHD, autism (AI adapts difficulty, pacing, and modality); (5) Speech — voice cloning restores communication for individuals who have lost speech (ALS, stroke) — Voiceitt learns non-standard speech patterns and translates to clear speech.

## Detailed Analysis
Sign language recognition (SLR): the Nature 2025 system uses a two-stage pipeline — (1) MediaPipe Holistic extracts hand landmarks, body pose, and facial expressions from video frames; (2) a transformer model processes the temporal sequence of landmarks, classifying gestures into ASL glosses (word-level signs) and generating English translations. The 47-article scoping review (2026) identifies three eras: (1) 2018-2020 — rule-based and simple ML models for single-task assistive tools; (2) 2020-2023 — deep learning for multi-modal assistive systems (speech-to-text + translation + object recognition); (3) 2023-2025 — LLM-powered adaptive systems that learn individual user patterns and preferences. Springer 2025 comprehensive review notes that AI accessibility faces unique challenges: (1) Bias — AI trained on able-bodied data may fail for users with atypical interaction patterns (tremors, speech impediments); (2) Privacy — assistive tools continuously capture audio/video of private environments; (3) Cost — state-of-the-art AI requires cloud connectivity and expensive hardware, creating a "digital accessibility divide." ResearchGate 2025 survey emphasizes the paradigm shift from "fixing the person" (medical model) to "adapting the system" (social model of disability) — AI is uniquely positioned to implement universal design at scale.

## Further Reading
- Microsoft Seeing AI App
- Google Project Euphonia (Speech Recognition for Impaired Speech)
- Voiceitt: Inclusive Voice AI for Non-Standard Speech