Specify the number of speakers

[a-kifumi]

Used in Japanese, the results are returned as if there were three speakers even though there are only two speakers.

I would like to limit the number of speakers by specifying options such as max_speaker.

[jjmaldonis]

Hey @yoshisansan! We do not currently have support for setting the max number of speakers, but it's likely that this will be supported in the future or the diarization results will improve so dramatically that it won't be needed as we continue to improve our models. Please keep an eye out for an email or updated docs with support for max_speakers. If/when that feature becomes available, we'll try to update this thread.

[a-kifumi]

I see. I look forward to it.

[erikmunson]

big +1 to this, I posted a note about why this is valuable on a different thread that seems to be a dupe of this one: https://github.com/orgs/deepgram/discussions/217#discussioncomment-7087013

[mehdigreefhorst]

+1 this is very needed, it often gets it wrong currently if you only have a single channel

[onurozgurozkan]

Is there any news about that feature?

[jjmaldonis]

Hey @onurozgurozkan, no ETA at the moment but I just sent your feedback to our team.

[arsi-rad]

+1 we find that diarization is unusable because it returns incorrect number of speakers, but we know how many speakers in our use case

[leafcutter-au]

+1 for me as well. My application would require a max speakers parameter as the ever growing number of speakers in the current transcription is problematic. Do we have an idea if this is still a planned feature?

[sthorat789]

+1 for me as well.