A voice for
every touch.
We build AI-powered assistive technology for the DeafBlind community — enabling true two-way conversation through Speech2Braille and Braille2Speech, running entirely offline on affordable hardware.
Breaking the
wall of isolation
DeafBlind individuals face profound isolation — unable to hear or see, yet possessing the full capacity for thought, feeling, and connection. AI4DeafBlind exists to dismantle that barrier using affordable, offline-capable AI that works in the real world — not just in a lab.
A fully bidirectional
conversation
Anyone can speak to a DeafBlind person — and any DeafBlind person can speak back.
- Fine-tuning speech recognition models and advancing the acoustic frontier for low-resource Indonesian language processing.
- Optimizing inference pipelines for edge deployment — squeezing Whisper and Piper into a Pi while keeping latency under one second.
- Curating training corpora from GigaSpeech2 and evaluating robustness across diverse speakers using Mozilla Common Voice ID v7.0.
- Word Error Rate — Whisper Tiny (ID)
- <20%
- Real-Time Factor on Raspberry Pi 5
- <1s
- Internet required — Fully offline
- 0G
- Target device cost — Pi 5 platform
- $150
From spoken word
to fingertip
A sighted person speaks Indonesian. In under one second, the words appear as Braille on the DeafBlind person's display — no internet, no cloud, no intermediary.
-
Speech Input
Spoken Indonesian via microphone
-
Whisper Tiny
Fine-tuned ASR, <20% WER, <1s RTF
-
Liblouis
Indonesian Braille translation
-
Braille Display
BRLTTY → any Braille display
From fingertip
to spoken word
A DeafBlind person types on their Perkins Keyboard. Their words are instantly spoken aloud — giving them a real voice in any conversation, without any special training required of listeners.
-
Perkins Key
Braille typed on 6-key Perkins input
-
Liblouis
Braille decoded to Indonesian text
-
Piper TTS
Neural text-to-speech on-device
-
Audio Output
Natural voice heard by the room
Testing with real users
in the field
Our first field deployment is with Helen Keller Indonesia at their school in Yogyakarta — putting our bidirectional smart speaker directly into the hands of the DeafBlind community.
Caregivers, teachers, and family members speak naturally in Indonesian — DeafBlind students read the words on their Braille display in real time. And when the DeafBlind student responds via their Perkins Keyboard, their words are spoken aloud for everyone to hear. A true conversation, at last.
Field Partner
Helen Keller Indonesia — YogyakartaDevice Finalization
Hardware assembly and software packaging on Raspberry Pi 5
Partner Briefing
Coordination with Helen Keller Indonesia's teaching staff and caregivers
Yogyakarta Field Test
Live deployment at the school — real users, real conversations, real impact
Evaluation & Iteration
Collect feedback, measure usability, refine the system for broader rollout
Open Source Release
Publishing code, Indonesian Braille table, and fine-tuned model weights
Sharing what
we discover
Technical write-ups on each pipeline — architecture, training data, benchmarks, and on-device performance.
Real-Time Indonesian ASR for DeafBlind Assistive Technology on Edge Devices
Braille2SpeechPerkins Keyboard to Natural Voice via Liblouis and Piper TTS on Raspberry Pi
Extending Liblouis with Indonesian Braille: Methodology and Open-Source Contribution
Open Source Publication · Braille Table · Liblouis · BRLTTY · Bidirectional Translation
Offline AI for Accessibility: Lessons from Deploying Bidirectional Speech–Braille in Yogyakarta
Disability Studies · Social Impact · Field Research · Helen Keller Indonesia
Low-Cost, High-Impact: The Case for Raspberry Pi in Disability Assistive Technology
Disability Publication · Affordability · Emerging Markets · Global South
Let's build
this together
Researchers, clinicians, disability advocates, and institutions — we welcome collaboration, feedback, and partnership to bring Speech2Braille and Braille2Speech to more communities around the world.
Get In Touch
