Recordings

An interview, a field recording or a filmed document is a source like any other. It sits on an item in your library, plays in the reader, and once it has a transcript Ask can quote it and cite the time.

Adding a recording

• Web: drop the file anywhere on the library. It is drafted as a recording, and Add details… in the tray lets you say what it is. To put it on an item you already have, use Add a copy on that item’s Read tab.
• iPhone and iPad: Add attachments on a reference offers Record audio, Videos from Photos and Documents, recordings or data (from Files). The library’s add menu has Record audio and Videos from Photos too, and files what you add as a reference of its own. A recording paused by a phone call carries on when the call ends.
• Mac: Import media from Photos, above the library list, takes videos as well as photographs. A video goes to the library as a recording.

Pangur takes MP3, M4A, AAC, WAV, FLAC, Ogg, Opus, AIFF and WMA audio, and MP4, M4V, MOV, WebM, MKV, AVI and OGV video.

Playing

A recording opens in the reader on the item’s Read tab. Audio shows its waveform: click anywhere on it to move through the recording. Video has captions once there is a transcript, picture in picture and full screen. Both have a playback speed control.

Some formats will not play in a browser or on an iPhone, an MKV or a WMA file among them. Pangur makes a copy that will, which takes a few minutes the first time, and says so while you wait. The file you added stays as it was.

Transcribing

A recording is transcribed only when you ask. Nothing is sent anywhere when you add it, record it or play it.

Until then the transcript panel reads “This recording has not been transcribed.” Under that it names the speech provider and model an administrator has chosen for this server, and choosing Transcribe sends the recording’s audio to that provider. For a video, where the server is set up to describe what is on screen, stills from it go to a second provider, which the panel names as well. Interviews are often recorded on terms that cover who may hear them, so check that the people recorded agreed to this before you choose Transcribe.

A long recording takes a few minutes, and you can leave the page while it runs. If it does not finish, the panel says why and offers Try again.

The monthly allowance

Each plan has a monthly allowance of transcription minutes. The transcript panel shows how many you have left this month, before you send anything. A recording longer than what is left is not sent, and the panel tells you its length and your remaining minutes. Correcting a transcript uses none.

Reading the transcript

The transcript is set out as dialogue, each speaker’s turn a paragraph with its time beside it. While the recording plays, the sentence being said is marked and kept in view. Click any sentence or time to play from there.

Where a video has been described, those passages appear in the pauses they belong to, labelled On screen. They are left out of the captions, which carry only what is said. Audio description in the player reads them aloud in the pauses, for anyone who cannot see the picture.

Speakers and corrections

When the model can tell speakers apart it labels them A, B and so on. Correct opens the transcript for editing: under Speakers, type a name over a letter and every passage by that speaker takes it. The text of each passage can be fixed in the same view. On iPhone and iPad, Speakers renames speakers; fix the words on the web.

Long recordings are transcribed half an hour at a time. Each part is given a short sample of every voice heard so far, so a speaker keeps the same letter from part to part. A voice that never spoke for more than a moment, or a fifth speaker, can still appear as “A (part 2)” where it may be someone already named; give both the same name.

Save correction stores your text as a new version. The panel then reads “Corrected, version 2”, counting up with each save. The transcript as the model made it, and every correction since, is kept; the newest version is the one Pangur shows, searches and cites.

Asking

A transcribed recording answers questions in Ask like any other text in the library. A passage from one is cited by time, such as 12:03–12:41, and the speaker’s name leads the passage quoted. Open at 12:03–12:41 under the answer opens the recording and starts it at that moment. What was seen in a video is quoted as “[On screen]”, so you can tell it from what was said.

A recording with no transcript has no text, so Ask cannot draw on it yet.

In an exhibition

The Audio and Video blocks of an exhibition take recordings from your library. Use the transcript copies the library’s transcript into the block: visitors read along, the passage being played is marked, and a click on any passage plays from there. Play from and Stop at show one stretch of a longer recording. The copy is taken when you bring it in, so after correcting the transcript choose Pull the transcript again.

Assistants and export

A connected assistant reads a transcript with get_transcript: timed passages with speakers, or one stretch of a long recording by start and end time. The tool never starts a transcription. If nobody has asked for one in the reader, it says so. See Assistant tools.

A project export holds each recording’s newest transcript as JSON, as captions and as plain text.