Text-to-speech

Finally, it is interesting to compare NETtalk s skills with the skills of commercially available text-to-speech systems such as DECtalk. Unlike NETtalk, DECtalk uses both extensive (and labor intensive) look up tables that store the phonetic transcription of common and not so common words and sets of phonological rules for words not in its main look up table. While DECtalk performs undeniably better than NETtalk, the comparison is not really a fair one. DECtalk is a combined r sult of many years worth of careful linguistic analysis and codification. NETtalk, on the other hand, is a self-contained. system that (in the span of a few hours worth of DEC VAX CPU time) can learn enough on its own from a set of simple examples to be both intelligible and accurate. 

Moulines and Charpentier, 1990] Moulines, E. and Charpentier, F. (1990). Pitch-synchronous waveform processing techniques for text-to-speech synthesis using diphones. Speech Communication, 9(5/6) 453-467. 

Allen, 1991] Allen, J. (1991). Overview of text-to-speech systems. In Furui, S. and Sondhi, M., editors, Advances in Speech Signal Processing, chapter 23, pages 741-790. Marcel Dekker. 

California Scientific Software s Brainmaker is a low-cost MS/DOS-based program for constructing multi-layer backpropagation networks based on several kinds of transfer functions. It comes with a set of trained networks whose tasks range from shape recognition to text-to-speech conversion. 

If only the larynx is involved, an externally applied artificial larynx can be used to generate a resonant column of air that can be modulated by other elements in the vocal tract. If other motor skills are intact, typing can be used to generate text, which in turn can be spoken via text-to-speech devices described above. And the rate of typing (either whole words or via coding) might be fast enough so that reasonable speech rates could be achieved. 

Technology transfer in the rehabilitation arena is difficult, due to the limited and fragmented market. Advances in rehabilitation engineering are often piggybacked onto advances in commercial electronics. For instance, the exciting developments in text-to-speech and speech-to-text devices mentioned above are being driven by the commercial marketplace, and not by the rehabilitation arena. But such developments will be welcomed by rehabilitation engineers no less. 

Text-to-speech systems have an enormous range of applications. Their first real use was in reading systems for the blind, where a system would read some text from a book and convert it into speech. These early systems of course sounded very mechanical, but their adoption by blind people was hardly surprising as the other options of reading braille or having a real person do the reading were often not possible. Today, quite sophisticated systems exist that facilitate human computer interaction for the blind, in which the TTS can help the user navigate around a windows system. 

Section 1.2. What should the goals of text-to-speech system development be ... 

Before delving into the details of how to perform text-to-speech conversion, we will first examine some of the fundamentals of communication in general. This chapter looks at the various ways in which people communicate and how communication varies depending on the situation and the means which are used. From this we can develop a general model which will then help us specify the text-to-speech problem more exactly in the following chapter. 

This chapter has introduced a model of communication and language that will serve as the basis for the next chapter, which looks at text-to-speech conversion in detail, and the subsequent chapters, which explore the various problems that occur and the techniques that have been developed to solve them. We should stress that what we have presented here is very much a model, that is a useful engineering framework in which to formulate, test and design text-to-speech systems. 

The above discussion shed hght on some of the issues of spoken and written cormnunication, and the particular process of reading aloud. From our understanding of these processes, we can now define a model of how text-to-speech conversion by computer can be carried out. The next section describes the model we adopt throughout the book, and the sections after that describe alternative models and how these compare. 

It is not strictly speaking necessary to start with text as input in creating synthetic speech. As we have clearly stated, our TTS model is really two processes, an analysis one followed by a synthesis one, and these are quite different in nature. As the analysis system will never be perfect, it is quite reasonable to ask whether there are situations where we can do away with this component and generate speech directly . Often this way of doing things is called concept-to-speech in contrast to text-to-speech. We shall stick with the generally used term Concept-to-speech , but we should point out that this can however mean a number of different things which we will now explain. 

In this section we identify the main challenges in high quality text-to-speech. First we describe four areas which we believe are the key problems that have been recognised as the main goals of TTS system building (but note, the problems are not always described in this way). Then we go on to describe two additional problems, which are starting to attract more attention and which we believe will be central to TTS research in the future. 

---