Abstract: Neural end-to-end text-to-speech (TTS) is superior to conventional statistical methods in many ways. However, the exposure bias problem, that arises from the mismatch between the training ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results