Labelling who speaks with Find speakers in Video by Zaweba
Transcript > Find speakers in Video by Zaweba groups words by who says them, on your PC with no script, and lets you rename each speaker everywhere.
A transcript of two or three people talking is hard to follow when it's one long run of words. You can't tell where the interviewer stops and the guest starts, or which panellist made the point you want to keep. Video by Zaweba adds speaker labels to the transcript with Transcript > Find speakers, so the text reads like a script with names.
What Find speakers does
Find speakers listens to the voices in the recording and groups the words by who says them. It runs on your PC, like the rest of the speech tools, and it doesn't need a script to do it. It works from the sound of the voices alone, so it's useful for unscripted footage such as interviews, meetings, podcasts and panel discussions.
The result is a transcript broken into sections, each headed by a speaker. At first the headings are generic, because the app can tell voices apart but can't know anyone's name.
Naming the speakers
Click a heading to rename that speaker, and the new name is used everywhere that speaker appears. Rename "Speaker 1" to "Priya" once, and every one of Priya's sections updates. It takes a few seconds per person and makes the transcript far easier to work with.
Why it helps when editing
With labels in place, editing an interview by text becomes much clearer. You can skim the guest's answers and skip the interviewer's questions, or find every point one panellist made. Deleting words in the transcript still cuts them from the edit, so you can trim one speaker's rambling answer without touching anyone else.
Saved transcripts are more useful too. A text file with names against each passage is ready for show notes, minutes of a recorded meeting or a written interview.
A note on accuracy
The README is open about one limit: it's untested on real same-sex voices. Two people of different sexes usually have clearly different voices and are easier to separate. Two men or two women with similar voices may sometimes be mixed up.
So check the labels, especially where two similar voices talk back and forth quickly. If a passage has been given to the wrong person, that's worth catching before you rely on it.
Clean voices, clean labels
Give each voice clean sound. Separate microphones help, and if two people were recorded on the left and right of one stereo file, breaking it out to mono clips first gives cleaner material for everything that follows.
Avoid crosstalk where you can. When people talk over each other, any system finds it hard to tell who said which word.
Run Find speakers after you've made the transcript and before you start heavy editing, so the labels are there to guide your cuts.
Rename everyone straight away, while you still remember who is who. A list of generic speaker headings is easy to muddle an hour later, especially when a recording has several guests who were introduced only once at the start.
From transcript to finished cut
Speaker labels sit on top of the transcript and text-based editing tools. Make the transcript, find speakers, name them, then edit by text. In Video by Zaweba the whole chain stays on your own computer, which matters when the conversation is private.
