- In the ZhiJun40 sentence corpus to do FA,zhijun_speakera2_30-fa-new, with the HW2000 sentence of MONOPHONE.MMF and data/list/ Mono.list as a standard fully trained model file, then perform FA
- FA is followed by the mono type of callout file, located below/gv/.../,
- Then, using the Ning script to get the full type of annotations,
- And then you use this set of annotations to make a composition.
- The result is an error:
=============== ang ================
ERROR [+2121] hinit:too Few observation sequences [0]
- There are not enough observations on syllable Ang.
- How to solve this problem: two ways:
- Reduce the number of HMM states, now is 5, reduced to 4 and 3 look.
- With ThinkPad to see, with HW2000 model file, after FA, zhijun40 sentence of the label file time callout information??
- At the same time, compare, before the use of zhijun40 sentence, self, first training to get the model file, and then the FA time standard information how??
8, in FA, with HW200 sentence monophone.mmf and data/list/mono.list as a fully trained model file, FA