How to get a transcript from a page that only plays the audio

Last updated 2026-09-17 · 8 min read

What is the short version?

You have a link to a programme, not a copy of it. Somewhere behind that page is an audio file, and there are usually several candidates published the same day under nearly identical titles — a bulletin, a repeat, and the interview you actually want to read.

1. Give Vizard Agent the page link and the date of broadcast.
2. Let it identify the recording before it transcribes anything.
3. Ask for the complete text, including the closing questions.

What do you need before you start?

The link, and one or two facts you can check the result against. Vizard Agent uses those facts to tell a three-hour morning programme from the forty-minute interview inside it, and to tell today's edition from yesterday's when both carry the same headline.

• The page link. The article or programme page.
• The date. The single most useful disambiguator.
• Who is on it. Presenters, guest, producer.
• Which part you want. The whole programme, or one interview.
• What you want back. Plain text, timecodes, speaker labels.

What do you type into Vizard Agent?

Name the date in the first message if you know it. Without it Vizard Agent has to infer which episode the page refers to, and a page that was updated after broadcast will happily point at a different recording than the one you were reading about.

Prompt

Variants worth knowing:

• "From 10 August." Settles which of several similar episodes.
• "The interview, not the bulletin." Names the segment inside the programme.
• "Including the last few questions." Guards against a truncated tail.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a public radio page whose player holds a forty-four minute party leader interview inside a much longer morning broadcast. The first recording it finds is the wrong one, and the recovery is the interesting part.

1. Reads the page and looks for a public audio reference behind the player.
2. Queries the broadcaster's own episode index rather than guessing a file name.
3. Transcribes a candidate — and it turns out to be the wrong item.
4. Narrows the search to the stated date and lists the editions published that day.
5. Verifies the candidate against the article's own details — the presenters and producer named in the text.
6. Compares two items from the same day on subject matter to choose between them.
7. Locates the full programme in the archive when the article's own reference will not resolve.
8. Cuts the interview out of the broadcast and uploads the isolated segment.
9. Transcribes the verified extract with timecodes and speakers.
10. Goes back for the final hundred seconds so no question or answer is missing.
11. Assembles one transcript from the main extract and the closing section.
12. Delivers it as a downloadable text file.

Step five is what turns a plausible match into a confirmed one. A title and a date are weak evidence when a broadcaster publishes six related items in a morning; the names of the people who made the programme are printed on the page and appear nowhere else, so Vizard Agent matches on those.

Step ten exists because long transcriptions are done in sections, and a section boundary is exactly where a question gets cut in half. Vizard Agent re-reads the end of the recording as a separate pass rather than assuming the last chunk ran to the finish.

Step three is a failure worth keeping visible. The wrong episode was transcribed in full before anyone noticed, which is precisely what happens when a link is treated as a file — and why the verification steps came after it.

What does the result look like?

One text file covering the whole interview, with timecodes and speaker labels, ending on the last answer rather than in the middle of a question. Vizard Agent tells you which recording it used and how it identified it, so you can check the provenance before quoting anything from it.

When does this not work well?

Some audio cannot be reached from a page at all. If the player streams from behind a login, a geographic restriction or a licence check, there is no public reference for Vizard Agent to resolve, and the honest answer is that you need to supply the file yourself.

• Login-protected players. No public audio reference.
• Region-locked broadcasts. The index resolves; the file does not.
• Live streams. Nothing to transcribe until it is published.
• Pages with no episode index behind them. Nothing to match against.
• Very old archives. The reference has often been retired.

How do you fix a result that came back wrong?

Tell Vizard Agent it has the wrong recording, and add whatever detail narrows it down — a date, a presenter's name, a phrase you remember hearing in the programme. That is a far faster correction than re-describing the whole request, because the transcription itself was never the problem.

• "Wrong audio, it was 10 August." The search is narrowed to that date.
• "That was the news, not the interview." The segment is cut out of the programme.
• "It stops mid-question." The closing section is transcribed and appended.
• "Who is speaking where?" Speaker labels are added to the text.

How does Vizard Agent compare to doing it yourself?

By hand you open the page, look for a download button that is not there, try to right-click the player, and eventually record the stream in real time. Forty-four minutes of interview is forty-four minutes of sitting there, and then the transcription starts.

Common questions

Can I just paste a link? Yes. Add the date as well, and Vizard Agent has far less to guess at.

What if the page has several audio items? Say which one. Otherwise Vizard Agent compares them against the article's subject.

Does it work for podcasts? Yes, and podcast pages are usually easier — the feed carries a direct reference.

Will it get speaker names? Yes, if you ask for them. Vizard Agent labels the speakers in the transcript.

How long can the recording be? Long. Vizard Agent transcribes in sections and stitches them into one file.

Why do transcripts lose the ending? Because the last section was assumed to run to the end. Vizard Agent fetches it separately.

Can I get timecodes? Yes, and they make the text far easier to check against the audio.

What about a language I do not read? Vizard Agent transcribes in the spoken language, and will translate if you ask.

Can it summarise it too? Yes, though ask for the full transcript as well so you can verify the summary.

What if I already have the file? Upload it and skip the search entirely — that path is much shorter.

Does it keep hesitations and repetitions? By default the transcript follows what was actually said.

How do I know it used the right recording? Vizard Agent tells you which item it matched, and on what evidence.

Can it do several episodes? Yes. Give Vizard Agent the links and it works through them one at a time.

Is the file downloadable? Yes, Vizard Agent uploads the finished text for you to download.