Update README.md

README.md

@@ -3,7 +3,7 @@ license: apache-2.0
 ---
 # Introduction
 
-The MossFormer2_SR_48K model weights for 48 kHz speech super-resolution in [ClearerVoice-Studio](https://github.com/modelscope/ClearerVoice-Studio/tree/main) repo.
+The MossFormer2_SR_48K model weights for 48 kHz speech super-resolution [1] provdied in [ClearerVoice-Studio](https://github.com/modelscope/ClearerVoice-Studio/tree/main) repo.
 
 This model is trained on large scale datasets inclduing open-sourced and private data.
 
@@ -64,4 +64,6 @@ myClearVoice(input_path='samples/scp/audio_samples_sr.scp', online_write=True, o
 ```
 
 Model Limitations: The current speech super-resolution model is trained on a clean speech dataset and is designed to work with clean speech inputs. For speech super-resolution on noisy speech audio, 
-we recommend using our 'MossFormer2_SE_48K' model for speech enhancement first, followed by 'MossFormer2_SR_48K' for speech super-resolution.
+we recommend using our 'MossFormer2_SE_48K' model for speech enhancement first, followed by 'MossFormer2_SR_48K' for speech super-resolution.
+
+[1] Shengkui Zhao, Kun Zhou, Zexu Pan, Yukun Ma, Chong Zhang, and Bin Ma, "HiFi-SR: A Unified Generative Transformer-Convolutional Adversarial Network for High-Fidelity Speech Super-Resolution", ICASSP 2025.