The recorded announcement on a jail call

Usually, a call placed through a correctional facility's phone system does not begin with the people on it. It usually begins with the system: a prerecorded voice that says where the call is from, that the call is recorded, and that the person answering is asked to accept it, often by pressing a key. Where the export includes those first moments, some of the recording has already gone by before anyone says hello.

Where the recogniser returns text for the announcement, that text is in the transcript. This guide is about what it looks like there, and what Pincite Audio does and does not do with it.

The script is not here, and why

Anyone looking for the wording of a jail call's recording wants the script. It is set by the facility and its phone vendor, it differs from one facility or vendor to the next, and it changes, so reproducing one version would be wrong for most readers and stale for the rest. What tends to hold is the shape: an identification of where the call is from, a sentence saying the call is recorded, and an instruction to the person answering. That shape is what a transcript can pick up, and it is enough to recognise the announcement when you meet it on the page.

Where it lands on a two-channel export

Many recorded-call systems, correctional ones among them, write each side of the call to its own channel, so an export can keep the caller and the person called apart. Whether this product treats a particular file that way is its own judgement, on uncalibrated thresholds: an ambiguous file is deliberately routed to the single-speaker path, where every line is attributed to one speaker and no channel labels appear at all. The rest of this section is about the case where the channels are treated as separate.

The announcement is inserted by the system rather than spoken by either party, so which channel carries it depends on how the system inserted it and how the export was made. It can land on one side, or on both, since both people may hear it; the product's own note says a match on both sides separates nothing.

That matters for one reason. Each line is labelled with the channel it came from, and those labels are designations — Channel 1 and Channel 2 — rather than names, until someone at your firm names the speaker. Which person is on which channel is not something the recording establishes. It is tempting to read the announcement as settling it: the message is played to the person who answered, so the channel carrying it must be theirs. That is an inference about the phone system and about the export's channel order, and nothing this product reads from the file confirms either. Pincite Audio does not draw it, and a reader should not draw it from the page without listening to the audio.

One case to know about: where the two sides overlap too much for the separation to be worth anything, the product withdraws the channel labels and the speaker lines fall back to a positional Speaker 1 and Speaker 2. The note and the markers for untranscribed spans go on naming channels, so a recording can carry channel numbers in those and none on its speaker lines. The audio is what connects the two.

What the transcript does with it

Where the recogniser returns text for it, the announcement's words come back as ordinary lines: timestamped, attributed to a speaker — the channel they arrived on, where the channels are treated as separate — and printed the same way as everything else. The recogniser is not told that a voice is a machine. There is no separate label for an announcement, and no line is dropped for being one.

On a recording whose declared kind is a jail call and whose two channels the product treats as separate, a note near the top of the transcript reports one observation: whether wording of the kind a facility announcement uses was transcribed in the opening stretch of the call, and on which channel. It leads with the fact that party roles on the channels are not identified, then reports that and nothing more. It never says which party is on which channel, and where wording matched it tells you to listen before naming either speaker. Where nothing was transcribed in that stretch, it says there was nothing to check. Where something was and none of it matched, it says that is not evidence there was no announcement, because, in the note's own words, the wording this checks for is uncalibrated and has never been checked against a real facility recording. That wording came from general knowledge of correctional telephony, and so did the description of the announcement on this page; both carry the same caveat.

That declared kind is chosen at upload and can be changed afterwards, from the recording's row on its matter page rather than from the transcript itself, until a transcript has been rendered from it or the recording is under an exclusion hold. So if a correctional call's note says nothing about an announcement, the declared kind and the channel separation are the first two things to check.

Where the note appears. On the transcript you read in the app and in the review pane. Not in the PDF, and not in what the connector hands to an assistant. What they carry instead is worth knowing before either leaves your firm. The connector withholds any speaker name nobody has attested, and while any speaker whose lines are in the transcript remains unnamed it attaches an advisory that speaker identities are machine-derived and unconfirmed and that no person's name should be attached to a speaker label — an advisory it stops attaching once every one of those speakers has been named. The PDF carries no speaker-attribution caveat anywhere, a deliberate decision rather than an oversight, so its designations have to be read for what they are.

Why it matters in review

The words are usually not anyone on the call speaking. In the opening stretch, a line saying that the call is from a correctional facility is usually the system talking, though a party can say the same words at any point. It prints under the channel's label because every line on that channel does, and a designation is a coordinate, not an attribution. Why a machine guess never gets a name is the same rule seen from the other side.

The line follows the label. When someone at your firm names a channel, the designation is replaced by that attestation and every line on the channel carries the name from then on — including the announcement's, where it landed there. Nothing on the page then separates the system's lines from the person's, so name a channel knowing its opening lines come along.

The clock can start before the conversation. Where the export includes the announcement and the acceptance prompt, a first exchange that begins some way into the recording is the shape such a call can have rather than evidence of missing audio, and a timestamp citation into that opening stretch is usually citing the system rather than a party.

The tones and the pause are not gaps. The key press that accepts the call and the pause while the system connects are not speech, and nothing is recorded for an ordinary pause, so no gap marker prints for one. A tone is a different matter: it is non-silent audio the source-side measurement cannot tell from a voice, so it can count toward a recording being held for review even where no speech is missing. The gap guide and the coverage guide cover both.

What this page does not say

Nothing here is about what the announcement means for the call, for the people on it, or for anyone who later relies on the recording. This page does not go there. It describes what the announcement is, where it appears, and what Pincite Audio does and does not do with it, which is the part a transcript can be checked against. Our terms of service set out what you are responsible for.