Accessibility
EchoSense-AI
See the World, Hear the Details
EchoSense-AI is an accessibility-focused visual assistant for blind and low-vision users. It describes surroundings, answers spoken questions about a scene, identifies currency, reads and translates text, and supports visual chat. Its core features can run privately on your phone with a downloaded model; optional online analysis is available only when you connect your own provider and give consent.
Five core tools, one tap apart
EchoSense-AI opens to Navigate and keeps Navigate, Currency, Read, Assistant, Translate, and Settings in one accessible bottom menu.
Navigate — understand your surroundings
Point your camera at a scene and EchoSense-AI describes visible objects, layout, people, and their relative positions. Ask a targeted question by voice, such as where your keys are, and have the current image analyzed for that request.
- Scene description using downloadable on-device multimodal AI or optional online analysis
- Voice commands for targeted object finding
- Optional Safety mode with object detection, spatial awareness, and proximity alerts
- Depth sensing through LiDAR on supported iPhones and ARCore Depth on supported Android devices
- When Safety is on, recent object and range observations can accompany the image as context for scene analysis
- Visual overlay showing detected object boundaries
Currency — identify banknotes
Point the camera at a banknote and EchoSense-AI reports the likely denomination aloud and on screen. With a downloaded Gemma 4 model, currency analysis remains available without an internet connection; users may also choose consented online analysis.
Read — access every word around you
A complete toolkit for getting text out of the physical world, whether that is a menu, a letter, a product label, or a full page of a document.
- Instant Text — continuous camera OCR that reads stable text aloud as you pan
- Guided Document Scan — spoken instructions help you align the page, then auto-capture with perspective correction and image enhancement
- Upload & Read — import images, PDFs, and text documents for extraction and reading
- Colour identification and light-level measurement
- Barcode and QR scanning, with links reported aloud rather than opened automatically
- Accessible Magnifier — full-screen zoom with freeze frame, high contrast, inverted colours, greyscale, and flashlight
- Text-recognition language selector, with auto-detect or a specific language for better accuracy
Assistant — a visual chat companion
Hold a natural, multi-turn conversation by text or voice. Share your camera view, attach an image or document, and ask follow-up questions — the assistant keeps the context of the conversation.
- Multi-turn text and voice chat with full conversation history
- Attach camera images, stored photos, PDFs, or text documents
- Switch between on-device AI and optional online AI using your own API key
- Streamed responses with readable formatting and speech output
- Configurable on-device, prefer-online, or online-fallback behavior
Translate — break down language barriers
Extract text from a sign, menu, camera image, PDF, or imported document and translate it into English. Translation can run with the downloaded on-device model; Android also supports downloadable offline language packs.
Your data stays yours
EchoSense-AI is built around on-device processing. No EchoSense-AI account is required to use its core features.
- The on-device model runs entirely on your phone — no images or prompts leave the device
- Optional online AI uses your own Gemini or OpenAI API key and requires explicit consent
- Before online scene analysis, EchoSense-AI identifies what will be sent and offers an on-device alternative
- Camera frames and analysis results are not intentionally retained after processing
- Conversation history stays in memory only while a chat is active
- Settings stay local, and API keys are held in the platform keychain or credential store
Powerful AI, completely offline
EchoSense-AI's offline visual intelligence comes from a downloadable model managed in Settings. No account is required to download the offered models, and each public model revision is pinned and verified before installation.
- Gemma 4 E2B — the smaller download, recommended for devices with at least 6 GB of memory
- Gemma 4 E4B — the larger, more capable option, recommended for devices with at least 8 GB of memory
- Advanced users can import a compatible local .litertlm model from Settings
When you need more power, you choose
For tasks that benefit from a cloud model, you can optionally connect your own Gemini or OpenAI API key. The app sends requests directly to the selected provider and provides consent controls and an on-device alternative.
- Connect Gemini or OpenAI with your own API key
- Scene-analysis confirmation identifies the image, prompt, and relevant object or depth observations that may be sent
- Choose the default behaviour — ask first, use online as a fallback, or prefer online when available
- Automatic fallback to on-device AI when the network is unavailable
Stay aware of your immediate surroundings
In Navigation mode, the Safety feature pairs object detection with depth sensing to warn you about nearby obstacles, announcing the object, its estimated distance, and its bearing.
- iPhone: uses the LiDAR depth sensor for metric distance estimation
- Android: uses the ARCore Depth API on supported devices
- Devices without depth hardware fall back to camera-relative proximity estimates
- A visual overlay marks detected obstacle boundaries
Safety and proximity features are experimental estimates. EchoSense-AI is not a certified mobility aid and should not replace a cane, guide dog, or other established safety practices.
Built for accessibility by design
Accessibility is the primary goal of every feature, not a layer added afterwards.
- Native labeled controls designed for VoiceOver, TalkBack, Apple Voice Control, and Android Voice Access
- Spoken feedback for every analysis result
- Voice input for targeted scene requests and Assistant messages
- Accessible Magnifier with zoom, freeze, high contrast, colour inversion, greyscale, and flashlight
- Large, high-contrast buttons with clear labels
- Customisable text overlay preferences
Compatibility & requirements
Android
- Android 12 (API 31) or newer
- ARM64 device (arm64-v8a)
- 6 GB RAM recommended
- Several GB of free storage for the model download
- Camera required for visual features
- An ARCore-supported device is recommended for depth-based Safety features
iOS
- iOS 17 or newer on an iPhone with a camera
- 6 GB memory recommended for Gemma 4 E2B; 8 GB for Gemma 4 E4B
- LiDAR-equipped models give the most accurate distance estimates in Safety
- Several GB of free storage for the model download