Abstract: This paper proposes a method for selecting training data for textto-speech (TTS) synthesis from dark data. TTS models are typically trained on high-quality speech corpora that cost much time ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results