Databases automatic labelling

While many possible approaches to statistical synthesis are possible, most work has focused on using hidden Markov models (HMMs). This along with the unit selection techniques of the next chapter are termed third generation techniques. This chapter gives a full introduction to these and explains how they can be used in synthesis. In addition we also show how these can be used to automatically label speech databases, which finds use in many areas of speech technology, including unit selection synthesis. Finally, we introduce some other statistical synthesis techniques. 

The use of automatic labelling algorithms is normally justified in terms of saving time [384], [5], [279]. In the past, we had small databases which could be labelled by hand, but as the databases we use today are larger, and because sometimes we wish to label the database very quickly, we require an automatic system. This argument basically says that automatic labelling is therefore a matter of convenience if we had an infinite supply of hand labellers we wouldn t have a problem. A second justification is that automatic systems can out perform human labellers in terms of accuracy and consistency, and so convenience alone is not the only justification automatic systems are in fact better, in our experience this certainly appears to be true. 

It is now realistic for even the smallest company to develop and manage some kind of database relatively cheaply. It is possible to hold a considerable amount of information on each individual customer, and it is relatively quick and easy to update the data and use the information automatically (for example, on invoices or address labels). 

When the labels are produced, the bar-code number is automatically generated and assigned to that container. The Receiving Section employee who requests the labels will input the quantify and number of containers received. The system stores this information in a tracking database. At the same time, a duplicate entry is automatically generated into a historical database that maintains this information as permanent. 

Finally, using a linear phonemic representation has the benefit that it makes automatic database labelling considerably easier. In Chapter 17 we shall consider the issue of labelling a speech database with phoneme units, both by hand and by computer. As we shall see, it is easier to perform this labelling if we assume a linear pronunciation model, as this works well with automatic techniques such as hidden Markov models. 

Two major problems stem from this. Firstly, any database which has been labelled with ToBI will have a significant amount of noise associated with the pitch accent label classes. Secondly, for any large scale machine learning or data driven approach, we need a considerable amount of labelled data to the extent that it is impractical to label data by hand. As we shall see in Chapters 15 and 16, virtually all other aspects of a modem data driven TTS system s data are labelled automatically, and so it is a significant drawback if the intonation component can not be labelled automatically as well. Because however the level of human labeller agreement is so low, it is very hard to train a system successfully on these labels we can hardly expect an automatic algorithm to perform better than a human at such a task. 

A method which determines whether a particular compound is already stored in the database or is new is an indispensable procedure for a compound-oriented chemical database. Such a method makes it possible to automatically assign a label, the registration number, to every compound in the database and, therefore, to identify the compounds unambiguously. Such a method is called compound registration. 

---