Hi,
I understand that we use different lf0, mgc and duration trees, because each one of them depends on different contextual factors. For example, when applying a certain question, the set of states will be split into 2 clusters and so on. Trees will be used for unseen models, so a model is given for unseen models. Regarding different trees .... -Do we have different questions for different features (lf0, mgc and duration )? where? -Having set of states (S); The result of state clustering (clusters: C1, C2 ,C3 ....) will be different clusters for each of lf0, mgc, duration (C1lfo, C2lf0, ... , C1mgc, C2mgc,...)?! or just (C1, C2, C3 ,...)? - How different trees will be used for unseen models |