Arabic(M) Speaker 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 =… See the full description on the dataset page: xvector embeddings.

Source: Hugging Face Hub (herwoww/arabic_xvector_embeddings). Metadata imported from the dataset’s Hub tags.