playdocs

This is an instrument with a document inside it. Point it at a URL and the page is pulled apart into blocks, read aloud in a voice from the DeltaVerse cast, and laid out as a waveform you can zoom to the individual sample — the way Audacity does it, because that is the only way to see whether a render actually did what it claimed.

What is around you

The border is a substrate: nine stations, eight of them on the ring, and each of the eight is lit by one band of the spectrum of the voice that is actually playing. The band edges are log-spaced across roughly 80 Hz to 8 kHz, because a voice does not live in the top half of a linear split and a linear split therefore lights only the two stations nearest the bottom.

The ground behind this text is the same voice drawn as three sub-expressions at once — the time-domain waveform, the spectrum, and the inflection, which is the rate of change of pitch and the thing a reader hears as delivery. It is turned down far enough to read over. It is always in exactly one of four declared states, and it never hides which. Live: an analyser is attached and carrying signal, and every pixel of it is measured. Resting: an analyser is attached and the audio is silent — a flat line with the faintest breath on it, because silence should look like silence and not like a waveform. Untapped: there is no analyser to read at all, and it draws a slow, deliberately synthetic standing wave that is shaped to look nothing like a voice. Unknown: an analyser is attached and reading it threw — a closed context, a node cut out of its graph — so there is no measurement this frame. That fourth state exists because the first three had no way to say I could not look, and the field went on reporting whatever it last saw; unknown is the absence of a measurement and is deliberately not the same word as silence.

On play and render the ground is following the element. On record it is following the microphone: arming the recorder attaches its analyser to this strip and to the ground, because a ground that stayed synthetic through a whole take while the strip beside it read live would be the untapped picture wearing a measured label. The one exception is stated rather than hidden: if the element’s own graph is already live, the microphone does not take either field away from what you are listening to.

Untapped is the state this page loads in. The audio graph is built on your first play, record or render — a context made without a gesture is created suspended and its analyser reads zeroes forever — so until you press something there is genuinely nothing to measure, and what is moving behind this paragraph is arithmetic rather than anyone speaking. The oscilloscope on the deck names the state it is in at all times, and while it reads untapped · generated the ground is not your voice and is not claiming to be. The contrast between that picture and a real one is the label: a decorative oscilloscope presented as a measurement is the one thing this page must never be.

The deck at the bottom holds two instruments. On the left, a live oscilloscope with an honest state on it: live when there is signal, resting when there is an analyser and silence, untapped when there is no analyser to read at all, and unknown when there is an analyser and reading it threw — which the badge writes out as unknown · not measured, because the absence of a measurement is not a measurement of silence. Beside the state the badge also names the picture whenever the picture is not the ordinary live one: frozen frame under reduced motion, no picture if voice-scope did not load and there is no oscilloscope there at all. The DreamKnob deck carries the same four states as lamps, from the same single value, so the badge and the lamp cannot disagree about the state. On the right, the waveform: a peak pyramid with a branching factor of 256, drawn as a minimum-to-maximum envelope with the RMS band over it, and below one sample per pixel it stops interpolating and starts drawing the samples themselves as dots.

What leaves this machine

To this host: the text you asked to have spoken, POSTed to the render service, and the GETs that read the audio back. The recorder is entirely local — a take never leaves the browser, and there is no upload path in the code to leave by.

To anyone else: nothing, until you point the bar above at a page on another host. Then your own browser fetches that page directly, and that host sees the request the same way it would if you had typed the address into its own bar — your address, your headers, the URL. That is the browser lane, and it is the default because it is the private one: none of it passes through this host, which never learns what you read. A server-side ingest lane would reverse the trade — that host would see us, and we would see what you asked for — and it is not deployed here. What comes back is parsed inert and reduced to text, so nothing on the far page can run here.

How to drive it

Press Render and the blocks below are sent to be spoken once; every play after that is the file. Drag on the waveform to select, and the selection can be looped, zoomed to, or used as the bounds of playback. The DreamKnob deck under the URL bar is the control surface — source, transport, level, meters, clocks, scope, voice and lamps — and the header buttons and side menus are the same controls as plain HTML, driving the same values, so the page is complete whether the deck loads or not.

The keyboard is listed in the ABOUT menu. Space plays, J and L skip, [ and ] zoom, F fits the whole take to the width.