Update README.md

README.md

@@ -16,4 +16,26 @@ configs:
   data_files:
   - split: validation
     path: data/validation-*
+license: mit
+language:
+- ar
+size_categories:
+- n<1K
 ---
+
+# 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](https://huggingface.co/datasets/arabic_speech_corpus) has 100 files for a single male speaker and [ClArTTS](https://www.clartts.com/) has 205 files for a single male speaker.
+
+The X-vectors were extracted using [this script](https://huggingface.co/mechanicalsea/speecht5-vc/blob/main/manifest/utils/prep_cmu_arctic_spkemb.py), which uses the `speechbrain/spkrec-xvect-voxceleb` model.
+
+Usage:
+
+```python
+from datasets import load_dataset
+
+embeddings_dataset = load_dataset("herwoww/arabic_xvect_embeddings", split="validation")
+speaker_embedding = torch.tensor(embeddings_dataset[1]["speaker_embeddings"]).unsqueeze(0)
+```