A drop-in replacement for the metadata of the RUSLAN Russian single-speaker
TTS corpus, with word-stress marks added to every multi-syllabic Russian
word in the transcripts. Audio is bundled unchanged.
The motivation is to train Russian TTS models (e.g. Kokoro, Tacotron, VITS,
StyleTTS, XTTS) that pronounce words with correct lexical stress.
Vanilla… See the full description on the dataset page:
https://huggingface.co/datasets/stilletto/ruslan-stressed.
