search

hitachi-speech / EEND

End-to-End Neural Diarization
MIT License
360 stars 57 forks source link

Have you tried this on the AMI dataset？ #18

Closed melttt closed 2 years ago

melttt commented 3 years ago

Great paper! I like the idea. but I get a terrible result in AMI dataset(about 40+ der)。

ChokJohn commented 2 years ago

@hshota0530 I also try to reproduce on AMI by myself, and I also get a terrible result. Could you share your pertrained model before adaption on AMI dataset or the final model on AMI dataset? Thanks!

shota-horiguchi commented 2 years ago

You may have seen our arxiv preprint (https://arxiv.org/abs/2106.10654), but the codes in this repo are based on our INTERSPEECH 2020 paper. We have some updates from the INTERSPEECH version, so results on AMI will be very different since AMI mostly consists of four-speaker conversations. Now we are planning to release our new implementation based on the arxiv version.

ChokJohn commented 2 years ago

Thanks for your reply. Is it convenient for you to tell me what you plan to release your update? You don’t seem to update this repository often.

You may have seen our arxiv preprint (https://arxiv.org/abs/2106.10654), but the codes in this repo are based on our INTERSPEECH 2020 paper. We have some updates from the INTERSPEECH version, so results on AMI will be very different since AMI mostly consists of four-speaker conversations. Now we are planning to release our new implementation based on the arxiv version.