简略的总结(mark)
https://github.com/CSTR-Edinburgh/merlin
We used two vocoders in these experiments: STRAIGHT [35] and WORLD [34]. STRAIGHT (C language version), which is not Open Source, was used to extract 60-dimensional Mel-Cepstral Coefficients (MCCs), 25 band aperiodicities (BAPs), and fundamental frequency on log scale (log F0) at 5 msec frame intervals. Similar, WORLD4 , which is Open Source, was also used to extract 60-dimensional MCCs, 5-dimensional BAPs, and log F0at 5 msec frame intervals.
features of neural networks thus consisted of MCCs, BAPs, and log F0 with their deltas and delta-deltas, plus a voiced/unvoiced binary feature. Before training, the input features were normalised using min-max to the range [0.01, 0.99] and output features were normalised to zero mean and unit variance.