HMM approaches

The HMM approach to phrase-break prediction was introduced by Taylor and Black [443] and has been extended by others [37], [369], In their basic system, Taylor and Black proposed that the decision on whether a phrase break should or should not be placed was dependent on two main factors ... 

In the HMM approach, we have a state for each type of juncture, so ignoring sentence breaks, we either have three states if we distinguish major and minor breaks or two states if we don t. Taking the second case first, we now have a state which gives ... 

More recently a number of approaches have been proposed which combine the advantages of decision tree approaches (use of heterogeneous features, robustness to curse of dimensionality) and the HMM approach (statistical, global optimal search of sequences). In addition, there has been somewhat of a re-awakening of use of syntactic features due to the provision of more robust parsers. Rather than attempt an explicit model of prosodic phrasing based on trying to map from the syntax tree, most of these approaches use the syntax information as additional features in a classifier [508], [209], [257]. 

HMM synthesis is an effective solution to the problem of how to map fi om the specification to the parameters. While most approaches aim to generate cepstral parameters, some generate formants and in this sense the HMM approach can be seen as a direct replacement for the provision of these rules by hand as described in Chapter 13. Issues still remain regarding the naturalness parameter-to-speeeh part of HMM synthesis, but as confidence gains in the ability to solve the specification-to-parameter part, new techniques will be developed to solve these naturalness problems. 

---