Consider rename the file "speaker-embedding.onnx"?
#1
by
pushdownLLM42
- opened
For pyannote/speaker-diarization-3.0
, the configuration file includes embedding: hbredin/wespeaker-voxceleb-resnet34-LM
, which indicates that the mode belongs to WeSpeakerPretrainedSpeakerEmbedding
.
However, if one wish to deploy the model offline, the name speaker-embedding.onnx
will confuse the function PretrainedSpeakerEmbedding
in pyannote/audio/pipelines/speaker_verification.py
and leads it to return PyannoteAudioPretrainedSpeakerEmbedding
, which is designed to load pytorch model instead of onnx model.
It might be better to include wespeaker in the model's name.