Context Map

What it is

Take the tiled Toronto map, remove the map, and keep the voice. What is left is three buttons and a running transcript. Press one and it tells you where you are and what is around you — the road you are on, the cross-street ahead, the notable places nearby — each with a distance and a clock-face or compass direction. There is nothing to look at and nothing to pan. The description is the map.

Why a map you don’t look at

Digital maps are, for the most part, not very accessible — and the parts that are can still be hard work to operate. Even the navigation aids built for blind and low-vision people tend to do one narrow thing well and say little about where you are and what your surroundings are like. That is getting better, but it is still not great.

A visual map is something a sighted reader takes in at a glance and a screen-reader user works through a great deal of structure to reach. For someone who only ever hears the map, all that structure — the tiles, the rotor, the focus management — sits between them and the one thing they came for: what is around me. The Context Map takes it away. The aim is to be as simple as it can possibly be: three big buttons, a line or a paragraph of plain speech, and no map-reading skill asked of anyone. You do not pan, zoom, or hunt through structure. You press a button and listen.

And because it speaks out loud rather than only feeding a screen reader, it is not only for screen-reader users. Someone with limited mobility who cannot easily turn to look around them — who cannot turn their head freely, or whose hands are busy with crutches or another mobility aid — can hold the phone and hear the same description.

The three controls

• Quick describe— one line, on demand: which way you face, the road you are on, the nearest main intersection, and the most worth-mentioning thing near you.
• Describe as I move— a running commentary as you walk; and because being turned around is movement too, it calls out when you turn, names your new direction, and re-casts everything around you to it. That is the part being tested in a crowd.
• Detailed surroundings— the full picture, read out and laid out as headings you can move through: ahead, to your right, behind you, to your left.

What it chooses to tell you

A description is only useful if it is the right handful of things. Each press gives you the road you are on; the nearest main intersection at each end of your block, the one ahead of you first; the notable named places nearby; and a one-line read on the character of the area you are in — each with a distance and a clock-face or compass direction.

What counts as worth mentioning depends on where you are. In a built-up area it leans on the things that matter on a city street — shops, places to eat, pedestrian crossings, transit stops. Out in the country those would leave you with almost nothing, so it speaks to the shape of the land instead: the lakes, ponds and wetland around you, woodland, and the nearest named settlement. The same press in the middle of Cambridge and beside a lake near Buckhorn should each feel like a fair account of that place.

Presence, never absence

The page only ever tells you what isthere. It never tells you something is absent — never “no crossing here”, never “nothing nearby”. The reason is the data: OpenStreetMap is mapped unevenly, and accessibility tags especially — dropped kerbs above all — are sparse. Silence means “not mapped”, which is not the same as “not there”. Treating a gap in the data as a fact about the world is exactly the kind of mistake that gets someone into trouble at a kerb, so the page refuses to make it.

Which way you’re facing

Everything is described relative to the way you are pointing: 12 o’clock is straight ahead, 3 o’clock is to your right. For that to be right, the page has to know your heading — and there are two honest sources for it, each trustworthy in different circumstances.

When you are moving, the most reliable heading is your course over the ground from GPS: it is immune to a miscalibrated compass, which on some phones can read fully half a turn out. When you are standing still, GPS has no course to give, and only the magnetometer — the device compass — knows which way you are turned. The design uses both, each where it is the trustworthy one.

For the current street trials the GPS side is switched offand the page runs on the compass alone, so we can shake out the magnetometer behaviour on its own. That is a temporary testing state, not the design — the two-source heading is built and waiting to be turned back on.

The compass came with a problem of its own. A phone’s reported compass heading assumes you are holding it flat in your palm; held upright, the way you naturally carry it while walking, the raw figure drifts as the phone tilts. iOS quietly corrects for this in the operating system; on Android it does not, so we compute a tilt-compensated heading ourselves from the phone’s full orientation — the true bearing of the back of the phone — which stays correct however the phone is held.

Being turned around is a kind of movement too. Rather than reacting to every small wobble, “Describe as I move” waits until a turn has settled, then announces your new direction and re-casts everything around you to it. That steadiness is the part being tested in a crowd.

Speaking alongside your screen reader

The page speaks through the browser’s Web Speech API. Where that is missing — on a hardened platform like GrapheneOS, for instance — it falls back to an ARIA live region instead. The two are never used together: a live region running alongside real speech would say everything twice.

Speech is the preferred channel, not merely the fallback’s alternative. ARIA live regions were not designed for a stream of frequent messages, and they have no way to stopan announcement part-way through — once one starts, you wait it out, even after you have already moved on. The Speech API can interrupt itself and pace what it says. It also opens the page up: because it speaks aloud in its own right, it works for sighted disabled users who do not run a screen reader at all.

Staying on while you move

While “Describe as I move” is running, the page holds a screen wake lock so the display does not sleep and cut the commentary off mid-walk. It tells you it is doing this: the screen will stay on until you switch apps or close the page.

That is as far as a web page can go. Keeping the audio running in the background— while you are in another app, or with the phone locked in your pocket — is not something a web page is permitted to do. It would need a native app, which is granted permissions the browser withholds. The wake lock is the ceiling of what is possible here, and the page is honest about that limit.

Built on OpenStreetMap

All of the place data comes from OpenStreetMap, the crowd-sourced map of the world. Building on it means its limitations come straight through to you: a building nobody has traced is missing; a shop that changed hands last year may still carry the old name. The Context Map can only ever be as current and as complete as the map underneath it.

The other side of that is the good part. OpenStreetMap holds a surprising amount of accessibility detail that most maps never surface — dropped kerbs, signal-controlled crossings, tactile paving, benches, accessible washrooms, ramps and stairs — and the Context Map brings it out and speaks it. That information has been in the map all along; it has rarely been read aloud.

What is frustrating is how much accessibility data sits locked inside commercial tools, where it cannot be contributed back to a shared map that anyone can build on — even when it was a volunteer who gathered it in the first place.

The map behind a page with no map

There is still a map underneath — you just never see it. The descriptions are not read off drawn tiles; they come from a search index. Every feature on the tiled Toronto map — every shop, crossing, bench and water’s edge — is also a record in an OpenSearch index, carrying its position, its kind, and its accessibility detail. It is the very same index that powers the search box on the visual map.

That is what lets this page exist without graphics at all. Instead of rendering a tile and asking you to interpret it, the Context Map asks the index a question — what is near this point, in which direction, how far — and reads the answer back. No tiles to draw, nothing to pan or zoom: the search data is the map.

A test, not a tool

This is unfinished, untested software, and it says so before you can use it. Every time you open it you have to read and accept a notice — that it can be wrong, that it is not for navigation or any safety decision, and that you use it entirely at your own risk. The notice appears on every visit, not once, because the software is unfinished and the stakes on a street are real. Keep using your usual ways of getting around at all times.

Where it can, it repairs quietly rather than alarming you. If the compass is slow to start, or a location request times out, it retries in the background instead of throwing an error into your ear in the middle of the pavement.

Try the interactive demo

Open the Context Map (opens in a new window)

You will be asked to read and accept the notice, then to allow location access. The demo takes over screen-reader announcements and runs on its own surface, which is why it opens in a new window. Close it to come back here.

Source

GPL-3.0. The Context Map is the spoken layer of the tiled Toronto map — it runs on the same engine and the same backend, and is reachable from within that map as well as on its own here — github.com/bobdodd/tiled-toronto-map. The place data is derived from OpenStreetMap, © OpenStreetMap contributors, under ODbL.