Hervé BREDIN comments

Results 270 comments of


                                            Hervé BREDIN

README Out of Date

I guess you tried with the develop branch of pyannote.audio, where this parameter has been renamed `token`. But it is still valid in official releases unless I missed something.

Optimize weighted embedding extraction with pyannote 3.1

> cc @hbredin, in case you can think of any other changes that could have caused this I can't think of any. I did not witness any change of performance...

Optimize weighted embedding extraction with pyannote 3.1

MPS support. `"linear"` interpolation is not yet supported with MPS backend.

segmentation/SSeRiouSS.py

Rev trained one here: https://huggingface.co/Revai/reverb-diarization-v2

fix(task): model can sometimes be None

Thanks. PR should be on develop branch. It says that there are conflicts. Can you please check?

fix(task): model can sometimes be None

Closing as I'd rather have `Model.from_pretrained` raise the error upstream.

Feat/joint diarization and embedding with prepared data

I just pushed a (possibly buggy) pipeline that seems to work with a joint model ```python from pyannote.audio.pipelines.speaker_diarization import SpeakerDiarizationV2 import torch device = torch.device('cuda') pipeline = SpeakerDiarizationV2('/path/to/joint.ckpt', batch_size=1, step=0.2).to(device)...

Hervé BREDIN

README Out of Date

Optimize weighted embedding extraction with pyannote 3.1

Optimize weighted embedding extraction with pyannote 3.1

segmentation/SSeRiouSS.py

fix(task): model can sometimes be None

fix(task): model can sometimes be None

Feat/joint diarization and embedding with prepared data

fix syntax warning for invalid escape sequence

Fix DDP training not working

feat: increase performance - only calculate embeddings on tensor channels which are not zero