4 ms·
No, this is just the vocoder part (like Wavenet) that converts from some representation to audio. You could probably use a Tacotron2-like approach to predict in
by woodson 8y ago
No, this is just the vocoder part (like Wavenet) that converts from some representation to audio. You could probably use a Tacotron2-like approach to predict input features (which appear to be 18 Bark-scale cepstral coefficients and two f0-related parameters) from text.