Audio search: find the exact second a word is spoken
Audio is a black box: you can't know what's inside without re-listening to it all. WeDispatch full-text audio search locates a word in the soundtrack, down to the second.
You skim text. You scroll, spot a word, jump to the right paragraph. Audio does not work that way. It is a continuous tape that only opens up when you play it from the start, and the bigger an audio library grows, the more it becomes a black box: you know a figure, a quote or a name was spoken somewhere, in one of your fifty voiced articles, but finding where is down to luck and a long re-listen. That is the problem audio search solves, and it is worth explaining how, because the way a tool goes about it decides how much you can trust it.
Searching sound is not the same as searching text
The easy temptation is to search the article's text and link back to the page. That beats nothing, but it stops at "it's in this article", and now you are in for eight minutes of listening to land on the three seconds you actually wanted. Written text and soundtrack do not even fully overlap: a word can be written and not spoken, or spoken differently.
WeDispatch audio search starts from the other end. Every audio is generated with word-level timestamps: a time grid that ties each spoken word to its start second. Searching a phrase then means locating it in that grid, not just in a body of text. The result is not "article number seven", it is "article number seven, at 4 minutes 12". One click drops you exactly there.
What it changes in practice
The gain is not a comfort feature, it is a change in the nature of the archive. Here is where it shows.
- A journalist checks a quote: they type the sentence, land on the second it is spoken, and confirm without replaying the whole interview.
- A community manager hunts the right clip to share: they find the passage, grab a timestamped link, and send it as is.
- An archivist reopens material from months ago: it becomes word-searchable again, instead of sitting there as opaque files.
In all three cases, what was impossible without a full re-listen becomes a few-second search. Audio stops being a dead end: it goes back to being material you can question.
Reliability comes from the text actually spoken
An audio search is only as good as the place it looks. Many tools lean on an approximate transcript, made after the fact, which invents words the voice never said. WeDispatch does not have that problem because it knows the exact text that was synthesised: search runs on what was really spoken, not on a reconstruction. It is the same foundation described on the French text to speech page, where the precision of the reading is the starting point for everything else.
And when a word does not appear as-is in the soundtrack, the tool says so, rather than sending you at random to a passage that has nothing to do with it. An honest "not found" beats a false positive: that is what separates a search you rely on from one you always double-check by ear.
What you need to use it
Search is scoped to your own content: it sweeps your account's audio library, and nothing else. It works on every audio generated since word-level timing arrived; an older audio becomes searchable again with a simple regeneration, no other handling required. That same timing also powers the highlighted text that follows the voice and the exportable subtitles: one synthesis job, several uses, which avoids paying twice for a reading you already produced.
What audio search does not do
For honesty, here are the limits. It does not understand meaning: it finds words and phrases, not reworded ideas. If a topic was discussed without the term you are searching being spoken, it will not guess it. It also does not search beyond your content, which is a decision rather than a gap: your archive is yours, and a search that went digging elsewhere would raise other questions.
What it does, it does to the second, on the real text, across your whole library. For a newsroom that builds on its archive, that is the difference between a pile of files and a library that answers questions. The full audio search page walks through the flow end to end.
Give your articles a voice with WeDispatch
This blog is itself voiced by WeDispatch. Curious how it sounds on your content?
Book a demo