Datasets:
Update dataset card: Correct license, add paper/code links, and expand tags (#1)
Browse files- Update dataset card: Correct license, add paper/code links, and expand tags (334c8f23a9b76d33971b0d3bdd90b8ec53803cb1)
Co-authored-by: Niels Rogge <[email protected]>
README.md
CHANGED
|
@@ -1,14 +1,22 @@
|
|
| 1 |
---
|
| 2 |
-
license: cc-by-nc-sa-4.0
|
| 3 |
language:
|
| 4 |
- ru
|
|
|
|
| 5 |
task_categories:
|
| 6 |
- text-to-speech
|
| 7 |
pretty_name: Balalaika
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 8 |
---
|
| 9 |
|
| 10 |
# A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models
|
| 11 |
|
|
|
|
|
|
|
|
|
|
| 12 |
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.
|
| 13 |
|
| 14 |
---
|
|
|
|
| 1 |
---
|
|
|
|
| 2 |
language:
|
| 3 |
- ru
|
| 4 |
+
license: cc-by-nc-nd-4.0
|
| 5 |
task_categories:
|
| 6 |
- text-to-speech
|
| 7 |
pretty_name: Balalaika
|
| 8 |
+
tags:
|
| 9 |
+
- russian
|
| 10 |
+
- speech-synthesis
|
| 11 |
+
- speech-enhancement
|
| 12 |
+
- audio
|
| 13 |
---
|
| 14 |
|
| 15 |
# A Data-Centric Framework for Addressing Phonetic and Prosodic Challenges in Russian Speech Generative Models
|
| 16 |
|
| 17 |
+
Paper: [https://huggingface.co/papers/2507.13563](https://huggingface.co/papers/2507.13563)
|
| 18 |
+
Code: [https://github.com/mtuciru/balalaika](https://github.com/mtuciru/balalaika)
|
| 19 |
+
|
| 20 |
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.
|
| 21 |
|
| 22 |
---
|