Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
The paper introduces Dialogs, a high-quality Russian conversational speech corpus designed for dialog assistants, comprising 20.6 hours of professionally recorded face-to-face dialogs segmented into 11,796 utterances. This dataset uniquely captures the nuances of turn-taking and expressive prosody, validated through crowd MOS tests that show superior expressiveness and naturalness compared to existing resources. Additionally, the authors demonstrate the corpus's utility by training a VITS2 model, confirming its effectiveness for expressive text-to-speech applications despite the limited data per speaker.
Dialogs achieves higher ratings for expressiveness and conversational naturalness than existing Russian speech datasets, setting a new standard for dialog assistant training.
We introduce Dialogs, a studio-quality Russian conversational speech corpus for dialog assistants. The dataset contains 20.6 hours of face-to-face acted dialogs recorded in a professional studio (44.1 kHz stereo) and segmented into 11,796 utterances across 3 speakers. Unlike read-speech resources, Dialogs captures turn-taking rhythm and expressive prosody, and provides per-utterance style/emotion labels spanning 12 categories. We validate corpus quality with crowd MOS tests, showing comparable audio quality and intelligibility to strong Russian studio baselines while achieving higher ratings for expressiveness and conversational naturalness. Finally, we train a VITS2 model as a proof of concept, demonstrating that Dialogs supports training expressive, dialog-like TTS despite limited per-speaker data.