I evaluated the HTS2.2 for Persian language and and one ambiguity and question is generated for me.
In HTS training when copying mono-phone HMMs to full context HMMs, and then HERest (embedded re-estimation) of full context HMMs are done, all question features are considered in full context phoneme classification. For example forward and backward distance in word, phrase , ... are considered ! My belief is these features are redundant that cause sparsity and so full context HMM re-estimated with only rare instances and so these low-training have straight affection in tying and clustering after that. I suppose only features about context phonemes must be considered in full context training of HMMs, and all features must be only considered in decision trees building. Is this saturation possible in HTK and HTS? And my thought is rational or not?