QHE wrote:
Hi, all: I trained a system with HTS, and find that there are a lot of duration models for states of 'sil', this waste many tree nodes and pdfs to store them. This is because my wav files has silence parts in the begining and ending, and the length are variable. Obviously, making complicated silence duration model is useless.
I wouldn't be so sure that a 'complex' silence duration model is useless.A context-dependent silence model might be a strange idea from a speech recognition point of view, but in synthesis it might be useful for predicting appropriate pause durations at different boundary types, for example.
You also need your silence model to be able to model silences of varying durations, because they occur in your training data. If your silence model is bad, you will get poor alignments between models and data during training, which would lead to worse results.
The answer to that is that you do need to get at least a little bit familiar with HTK, because the HHEd command you want can be found in HTKBook.So I want to tie the states of 'sil', making the duration fixed. But I'm not familiar with HTK, could anyone give me a hint how and where to write the hhed command?
Simon -- The University of Edinburgh is a charitable body, registered in Scotland, with registration number SC005336.