Read the paper "TransForm Mapping Using Shared decision Tree Context Clustering for hmm-based cross-lingual Speech Synthesis" (2)

3 cross-lingualspeakeradaptationusing STC with a bilingual corpus

First paragraph:

2. In the state mapping technique described in the previous section, the mismatch of language characteristics affects the map Ping performance of transformation matrices because only the acoustic features is taken into account in the kld-based map -Ping construction. To improve the mapping performance, we do not have only acoustic features if also contextual factors when constructing the TR Ansform mapping.
   1. These two words were read yesterday,
   2. Temporarily do not go too buckle details, you can take a little bit of depth of understanding each sentence, but do not buckle too deep can
   3. Plainly, this paper and Yijian Wu's paper are all done by state maping,
      1. Yijian Wu is based on KLD, only considering the acoutic features
      2. This paper is based on STC, taking into account ACOUSTC features and contextual factors
4. by using contextual factors, we can also take articulation manners and suprasegmental features to account for the MA Pping construction.
   1. How to pronounce
   2. Super-Sonic segment features
      1. What is a feature of a super-sonic segment???
      2. It's been a while, and it's not clear what the features of the super-Sonic segment are.
6. In this section, we propose a novel transform mapping technique based on GKFX decision tree Context Clustering (STC) for Cross-lingual speaker adaptation, which can reduce the influences of the speaker and language mismatches between average Voice models of input and output languages.
   1. This paper presents a novel State mapping technology, based on STC Tree
   2. Wipe, I remember. The STC tree is not the author's original, has been proposed for the SAT process, in the adaptive, if there are multiple speakers, and the gap between the speaker is very large, with STC can be trained a better average voice model
   3. However, the author is also very strong, because, before the state mapping is based on kld, he can use the SAT technical STC to do state mapping is also very powerful
      1. First of all, he's very clear about state mapping.
      2. Then, clearly understand what kld state mapping's weakness is.
      3. Clearly understand the whole STC process.
      4. Clearly understand the advantages of STC can overcome the shortcomings of kld.

Read the paper "TransForm Mapping Using Shared decision Tree Context Clustering for hmm-based cross-lingual Speech Synthesis" (2)