Improve dataset card: Update license, add explicit links, and enhance tags and structure

This PR improves the dataset card for the Balalaika dataset by:

- Correcting the license in the YAML metadata to `cc-by-nc-nd-4.0` to accurately reflect the dataset's license stated in the card content.
- Adding an explicit link to the associated paper on Hugging Face Papers (`https://huggingface.co/papers/2507.13563`).
- Adding a direct link to the GitHub repository (`https://github.com/mtuciru/balalaika`) for easier access to the code.
- Adding additional relevant tags to the metadata (`russian`, `audio`, `speech-synthesis`, `speech-enhancement`) for better discoverability.
- Restructuring the top section of the dataset card for better readability and information hierarchy, including an explicit "Abstract" section.

README.md

@@ -1,13 +1,27 @@
 ---
-license: cc-by-nc-sa-4.0
 language:
 - ru
+license: cc-by-nc-nd-4.0
 task_categories:
 - text-to-speech
 pretty_name: Balalaika
+tags:
+- russian
+- audio
+- speech-synthesis
+- speech-enhancement
 ---
 
-# A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models
+# Balalaika Dataset
+
+This repository contains the **Balalaika** dataset, a novel dataset comprising more than 2,000 hours of studio-quality Russian speech with comprehensive textual annotations, designed for addressing phonetic and prosodic challenges in Russian speech generative models.
+
+**Paper:** [A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models](https://huggingface.co/papers/2507.13563)
+**Code:** [GitHub Repository](https://github.com/mtuciru/balalaika)
+
+---
+
+## Abstract
 
 Russian speech synthesis presents distinctive challenges, including vowel reduction, consonant devoicing, variable stress patterns, homograph ambiguity, and unnatural intonation. This paper introduces Balalaika, a novel dataset comprising more than 2,000 hours of studio-quality Russian speech with comprehensive textual annotations, including punctuation and stress markings. Experimental results show that models trained on Balalaika significantly outperform those trained on existing datasets in both speech synthesis and enhancement tasks.
 
@@ -286,7 +300,7 @@ Supported models:
 ## Citation
 
 If you use this pipeline in your research or production, please cite:
-```
+```bibtex
 @misc{borodin2025datacentricframeworkaddressingphonetic,
       title={A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models}, 
       author={Kirill Borodin and Nikita Vasiliev and Vasiliy Kudryavtsev and Maxim Maslov and Mikhail Gorodnichev and Oleg Rogov and Grach Mkrtchian},