How to use this model for Speech Synthesis about qppwg HOT 2 CLOSED

Hi @Piteryo,
QPPWG needs the F0 values, but the F0 values can be obtained from input speech/audio or an F0 predictor.
For many conventional TTS systems, the F0 values are predicted using an F0 predictor.

However, the current SOTA TTS models such as Tacotron predict only mel-spectrograms, which entangle F0, harmonic, and spectral information.
Therefore, if you want to use QPPWG for them, you need to develop an extra F0 predictor or disentangle the F0 information from the predicted mel-spectrograms.

from qppwg.

@bigpon Thank you a lot. Everything is clear now.

from qppwg.