Read a recording instead of scrubbing through it

Finding one sentence in an hour of audio means dragging a slider until a voice starts to sound familiar. The slider speaks in minutes, the memory holds words, and neither one can read the other.

Text doesn't have this problem.

An article gets skimmed, searched, quoted and left mid-paragraph for later. A recording plays at the speed it was spoken, from wherever the slider lands, and a second pass costs the hour all over again.

The speeds are lopsided too. Conversational English runs near 150 words a minute, so an hour of talk comes to roughly 9,000 words, and 9,000 words is an article-length read that skimming cuts down further.

The upload page now takes audio. An MP3 or M4A dropped at readplace.com/upload comes back as a transcript in the Readplace reader, written as paragraphs stamped with the minute and second they were spoken, with the recording kept on the same page.

Every paragraph knows its second, so reading and listening can trade places mid-thought.

What the upload takes

MP3 and M4A files go up to 256 MB each, on the same page that takes DRM-free EPUB books. A file Readplace can't read as audio gets a notice asking for one that plays on the device, and a video file is refused as an unsupported type.

No account is needed to start. The upload and the review both work signed out, and Save to my library is the step that asks for a sign-in, then returns to the same review.

A second stamped on every paragraph

Each stretch of speech becomes its own paragraph, and each paragraph opens with the moment it starts, written as 02:41 or 1:07:35. A heading lands every 5 minutes, so an hour of talk reads as 12 short blocks, not one wall of text.

That structure is what skimming needs. A claim half-remembered from somewhere in the middle is a scroll away, and the scroll ends on the exact second to hear it back.

The language is detected from the recording, not chosen in a setting.

TBH, the exact part is the stamps, not the words: the transcription is machine-made, so an uncommon name can come out wrong, and nothing in it tells one speaker from another.

The recording stays on the page

A tap on a paragraph's stamp plays the recording from that line.

While it plays, the paragraph being spoken stays marked, so a reader who drifted ahead can find the voice again. The page address keeps the second as well, and a reload lands back on the same moment.

Skimming ahead of the playback is the point. A flat paragraph gets passed at reading speed. The one where a lecturer slows down, or a guest finally gives the number, earns a tap and plays in the speaker's own voice.

Minutes measured before any are spent

Transcription draws on a daily allowance of transcript minutes, and the upload page states the day's figure. The review shows the recording's length and how much of the allowance it would use. The minutes come off when the file is saved, not when it's uploaded.

The review rounds up to whole minutes, so a 90-second memo shows 2, but the meter runs to the second: only the minute and a half comes off, and the other 30 seconds stay in the day's allowance.

A recording longer than the whole day's allowance is turned back with a note to split it, instead of being transcribed up to a cut and stopped. The allowance resets at 00:00 UTC.

A transcript only its uploader can open

A saved article in Readplace is keyed by its URL and shared by every account that saves the same URL. An upload has no URL, so it gets a private key of its own, and the transcript belongs to the account that saved it. The reader offers it no share link.

A recording with no speech in it says so rather than producing an empty page: "No speech was found in this file, so there's no transcript." A transcription that fails says so too, and the file can go in again.

The file that's been sitting there

There's usually one nearby. A voice memo from a meeting that ran long, or the episode downloaded for a flight and half finished, waiting on a free hour that keeps not arriving.

A first upload at readplace.com/upload settles whether the scrub bar gets missed. The readlist files the result next to the articles already on it.