Update README.md

README.md

@@ -4049,12 +4049,13 @@ language:
  It is pre-trained on over 1 million hours of publicly available speech datasets. 
  It requires fine-tuning to be used in downstream tasks such as Speech Recognition or Translation. 
  Its hidden states can also be used with k-means for semantic Speech Tokenization. 
- XEUS uses the [E-Branchformer]() architecture and is trained using [HuBERT]()-style masked prediction of discrete speech tokens extracted from [WavLabLM](). 
+ XEUS uses the [E-Branchformer](https://arxiv.org/abs/2210.00077) architecture and is trained using [HuBERT](https://arxiv.org/pdf/2106.07447)-style masked prediction of discrete speech tokens extracted from [WavLabLM](https://arxiv.org/abs/2309.15317). 
  During training, the input speech is also augmented with acoustic noise and reverberation, making XEUS more robust. The total model size is 577M parameters.
 
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/630438615c70c21d0eae6613/BBRKYvTjJmx2B5oyWBLcZ.png)
  
- XEUS tops the [ML-SUPERB](https://arxiv.org/abs/2305.10615) multilingual speech recognition leaderboard, outperforming [MMS](https://arxiv.org/abs/2305.13516), [w2v-BERT 2.0](https://arxiv.org/abs/2312.05187), and [XLS-R](https://arxiv.org/abs/2111.09296). XEUS also sets a new state-of-the-art on 4 tasks in the monolingual [SUPERB]() benchmark.
+ XEUS tops the [ML-SUPERB](https://arxiv.org/abs/2305.10615) multilingual speech recognition leaderboard, outperforming [MMS](https://arxiv.org/abs/2305.13516), [w2v-BERT 2.0](https://arxiv.org/abs/2312.05187), and [XLS-R](https://arxiv.org/abs/2111.09296). 
+ XEUS also sets a new state-of-the-art on 4 tasks in the monolingual [SUPERB](https://superbbenchmark.org/) benchmark.
 
  More information about XEUS, including ***download links for our crawled 4000-language dataset***, can be found in the [project page](https://www.wavlab.org/activities/2024/xeus/) and [paper](https://wanchichen.github.io/pdf/xeus.pdf).