Upload call transcripts
Create a transcripts collection and upload interview or call transcripts as text, SRT or WebVTT files for Adduce to read.
A transcripts collection is a source with no credential: you upload the files yourself, and Adduce reads what the people in them said.
Before you begin
- Role: project Editor or Lead. A workspace Admin is a lead on every project.
- To hand: transcripts as
.txt(or.text),.srtor.vttfiles, each under 5 MB and 10 MB in total per upload. - Time: two minutes, plus a few minutes for Adduce to read the files.
What Adduce reads
| Surface | What it becomes |
|---|---|
| What each person said | Evidence quoting them, attributed to the speaker the file names |
| Speaker labels | The speaker on each quote — Name: text lines in a .txt, or <v Name> in a .vtt |
Timings in .srt and .vtt cues | Where in the call the quote came from |
Nothing leaves your workspace: the files stay in the storage Adduce owns for your organisation, and you can delete any of them.
Permissions
None. This is the one source that connects to nothing, so there is no credential, no scope and no connection test.
Steps
In Adduce, open the project, choose Sources, then Add source, and choose the Transcripts tile.
The upload flow has three steps rather than four, because there is nothing to connect to. On Name the collection, fill both the Collection name — what your team will call this set of calls — and the Source name it came from, then choose Create.
Open the new collection and choose Upload transcripts. Pick up to 25 files at a time, 5 MB per file and 10 MB in total. A file that is not .txt, .text, .srt or .vtt is refused before anything is sent, with a message naming it. An upload over either size limit is refused whole and no file is named; an oversized file is reported as Invalid or oversized multipart upload.
Each file appears in the table straight away as Queued, then moves to Processing, then Extracted once its evidence has been written. A file Adduce could not read shows Failed. A file’s own page is finer: while the file is being read it shows Parsing, then Extracting.
Open a file to see what Adduce made of it: its Format, its Words, a signal count — the evidence taken from it — and its Speakers.
Speaker detection
How Adduce finds who said what depends on the format. A .vtt file's <v Name> voice tags are used directly. An .srt file gives timings but usually no names. A .txt file is read line by line, and a line beginning Name: is treated as that person speaking. Files with no speakers still produce evidence — it is simply unattributed.
Deleting files and collections
Delete a single file from the row it sits in, or delete the whole collection from the ⋯ menu on its page by choosing Delete source and typing the collection's name. Either way the evidence taken from what you deleted goes with it.
If it did not work
Why was my file refused before it uploaded?
The extension is not one of .txt, .text, .srt or .vtt, and the message names the file. A file over 5 MB, or an upload over 10 MB in all, is refused by the server instead, whole and without naming a file. Export the transcript again in one of those formats; 5 MB is far larger than a normal transcript, so a file over it is usually a video or an archive by mistake.
Why was a file skipped?
The project already has a transcript with exactly the same contents, so Adduce did not store it again — counting the same call twice would make its evidence look stronger than it is. The message names the file it matches. The check covers every transcripts collection in the project, and two identical files in one upload count too: the first is kept. A file with even one character changed is treated as new. If the earlier copy failed, retry it from the file list instead; to upload the same file again, delete the earlier transcript first.
Why is a file stuck on Queued?
Reading a file is queued work, not something that happens while you wait. If it stays queued much longer than the others in the same upload, delete it and upload it again.
Why does a file show Failed?
Adduce could not read it — most often an .srt or .vtt whose cues are malformed, or a file that is empty. Open it in a text editor to check, then upload the corrected file.