Who said what

Take the speaker out of a transcript and a claim turns into a rumor - present in the text, attached to no one. It reads the same whether the host said it, a guest said it, or someone read it off a slide.

That is a real gap, not a cosmetic one. Who said something changes what it is worth: an offhand line from a guest is not the same fact as a considered claim from the host, even if the words on the page are identical.

Vocateca runs speaker diarization on-device, on by default. It separates the recording into speaker turns and labels them in the output, so the transcript shows who said each line, not just what the line was.

That labeling happens at the same time as the transcription, not as a separate pass you have to remember to run. A two-person interview or a panel with five voices comes out with turns already split by speaker.

Once the speaker labels exist, a quote can be attributed to whoever said it, not just to the episode it came from. That distinction matters the moment you are citing a claim in a post or checking who actually made an argument in a recording with more than one voice.

An unlabeled transcript can tell you what was said in a show. A labeled one can tell you what a specific person said, which is the version you can actually cite.

Diarization separates voices and assigns labels; it does not guarantee every label is right on every line, especially with overlapping speech or a voice that only appears for a sentence. Treat labels as a strong default, not a claim of perfect accuracy.

Even with that caveat, a transcript with speaker turns beats one without them. Attributable and mostly right is more useful than anonymous and technically complete.