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.