Change a label in audio.manual recipe

,
1

Is it possible to swap a label in the audio.manual recipe? For instance, to swap an annotation's SPEAKER_1 label to a SPEAKER_2 label?

For instance, I've already used pyannote-audio to automatically annotate speakers. The start and end of the speaker labels are often correct, however, the speaker assignment is more often incorrect. It would be easiest to be able change the label of an annotation to another speaker, rather than deleting and re-creating the speaker annotation from scratch each time.

image

Thanks for your help.

2

Thanks, this is a really good point! At the moment, we don't have a "selected" state of audio segment, so there's no way to select a region and then change the label (which is how it currently works in the image_manual UI). I think it'd be a very useful feature because re-annotating a region is obviously very inconvenient.

I will add this to my list of enhancements!

3

Cool thanks! That would be great :slight_smile:

4

Another feature that would be very useful for audio and video annotation is being able to increase the speed of playback, e.g. 1.25x, 1.5x, 2x etc.

5

We currently have this available as the audio_rate setting, which you can define in the config on startup (and which will then apply to all playbacks): Annotation interfaces · Prodigy · An annotation tool for AI, Machine Learning & NLP

It could be nice to also expose this as a setting to toggle at runtime – although, we're always conscious of keeping the UI straightforward and avoiding too many distracting runtime settings and toggles.

6

Thanks Ines, I didn't know about that.

I think it would be useful to have that as a UI setting (or keyboard shortcut), so that you can listen quickly to find the specific part of the audio to annotate and then slow it down when doing the actual annotation. It also gets harder to tell the difference between different speakers when the audio is sped up so being able to change the speed at will would be useful in that sense.