---
pretty_name: 'WitchSpeech: Russian voice lines Witcher 3 TTS dataset'
task_categories:
- text-to-speech
language:
- ru
---
# WitchSpeech: Russian voice lines Witcher 3 TTS dataset

This is a repack of the dataset
[so-vits-svc-4.0-ru-The_Witcher_3_Wild_Hunt](https://huggingface.co/Rootreck/so-vits-svc-4.0-ru-The_Witcher_3_Wild_Hunt) for TTS purposes.    
Unlike the original dataset, this dataset also contains transcriptions for voice lines.    
Transcriptions include stresses for words, even for words with just a single vowel.    
Additionally, there is a `metadata_source.csv` file, that contains voice lines text “as-is”.

Dataset info (see details in `stats.txt`):  
     
**Sample rate**: 48 000    
**Total time**: 21.49    
**Number of speakers**: 34    

In addition to the text, there are also non-speech sound tags in `metadata.csv`    
(see `non_speech_tags.txt` for details).
How `metadata.csv` was created:
1. Russian voice lines texts were extracted from the game and matched with audio samples    
from [so-vits-svc-4.0-ru-The_Witcher_3_Wild_Hunt](https://huggingface.co/Rootreck/so-vits-svc-4.0-ru-The_Witcher_3_Wild_Hunt) dataset.
2. Stresses for words were automatically added with [RUAccent](https://github.com/Den4ikAI/ruaccent) library.    
3. Non-speech sound tags were added from the original voice lines text.

`metadata.csv` structure:    
path to audio | text | speaker id (see `speaker_ids.json` file for details)

This dataset was made possible by the tools and resources created by the following people:    
[Rootreck](https://huggingface.co/Rootreck)    
[Den4ikA](https://github.com/Den4ikAI)