herwoww/arabic_xvector_embeddings

Arabic Speaker Embeddings extracted from ASC and ClArTTS

There is one speaker embedding for each utterance in the validation set of both datasets. The speaker embeddings are 512-element X-vectors.

Arabic Speech Corpus has 100 files for a single male speaker and ClArTTS has 205 files for a single male speaker.

The X-vectors were extracted using this script, which uses the speechbrain/spkrec-xvect-voxceleb model.

Usage:

from datasets import load_dataset

embeddings_dataset = load_dataset("herwoww/arabic_xvector_embeddings", split="validation")
speaker_embedding = torch.tensor(embeddings_dataset[1]["speaker_embeddings"]).unsqueeze(0)