Is there a library available where a person could use this algorithm for text-to-speech? What is necessary to generate the input features for this algorithm?
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.
2 comments
[ 3.3 ms ] story [ 14.3 ms ] thread