Share use cases for speaker diarization / speaker identification

[raivisdejus]

Many people have requested speaker identification feature where Buzz would add some identifiers to the transcripts to know what sentences where said by what person. Technically speaker diarization is quite easy, there are several tools to help with this.

In this discussion I would like to collect community feedback on how you plan to use it. Please share your full workflow. Where the audios will come from and how you want to use them afterwards? What possibly imaginary features may help you in the process. These use cases are important to figure out the best way to integrate speaker identification in the Buzz UI and possibly add extra tooling.

I myself can imagine the following scenario / workflow.

I am a journalist and I have recorded an interview among three people. I will need to write a magazine article based on the interview, so I would import the audio file from my recorder into the Buzz to transcribe. I would need Buzz to identify the speakers in the transcripts and add labels to each transcript of who said it. Audio transcript with identified speaker labels will be copied to a text document for further editing of the article.

Imaginary features for Buzz to have:

• Setting to specify number of speakers to improve identification quality. If number of speakers is known this setting can be passed to the identification process, if it is not passed Buzz should do its best to identify the speakers and guess how many of them are in the audio
• After speaker identification it would be nice to preview speech from each identified speaker and specify a label in some text box, to tell that SPEAKER01 is John Smith

If you know of some good speaker identification related features in some other application that you would like to see in Buzz, please share those.

[marrie]

I like the idea of speaker 1 and speaker 2 etc. I also like the idea of a label you can apply if the speaker is known.
If I'm doing an interview or if I'm transcribing a video or class, that might also be a very valid use case.

[Abandersen04]

I'm new in this (github) area and just a happy user of Buzz. My case will be similiar to yours. I dont now any specific tools, but ChatGPT for instance is pretty good at identify who's asking and who answering.