49.1 hours of filtered Russian speech recordings derived from the 1,300-hour GOLOS corpus. The dataset consists of audio segments processed through the BALALAIKA pipeline specifically for generative speech modeling.
Use Cases
- Train text-to-speech (TTS) models using the filtered audio segments and Russian transcriptions
- Fine-tune speech synthesis systems using the high-quality annotations from the BALALAIKA pipeline
- Evaluate Russian speech generation performance on a curated subset of the GOLOS corpus
Strengths
- 49.1 hours of curated Russian speech audio
- Filtered from a raw source of over 1,300 hours
- Processed using the BALALAIKA annotation pipeline
- Licensed under CC BY-SA 4.0