DailyTalk was developed using the open-source dialogue dataset DailyDialog and the FastSpeech framework, recording and modifying more than 2,500 dialogues from DailyDialog.
The researchers identified the limitations of conventional TTS models related to context representations that overlooked the importance of dialogue, background noises, and recording quality. DailyTalk’s high-quality dialogue speech dataset for TTS systems analyzes both general as well as conversational speech synethsis quality.