Font Size: a A A

Research And Implementation Of Speech Synthesis System For Spaceflight Testing

Posted on:2008-11-28Degree:MasterType:Thesis
Country:ChinaCandidate:J C DangFull Text:PDF
GTID:2178360242976199Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
This paper reviews the history of Text to Speech (TTS) system, its applications and key technologies involved, and then, a detailed explanation of a TTS system which is designed for spaceflight testing applications are presented.The common sense of the TTS definition can be regarded as a computer-based application system which takes text as input and outputs speech wave according to the inputs. TTS can analysis the input text automatically and decide the pronunciations, break and tone, then synthesize the speech wave by pre-stored speech units or its features to play out as output.A typical commercial TTS system always has a complex structure, a large size and sufficient to deal with ordinary text. However, when a piece of text from a professional domain is taken as input, the synthesis result will be degraded. This is due to the glossary contained in the text which is rarely found in the common lexicon. The errors of combination of words and placement of breaks will be stemmed from mismatch of words and the speech quality will be degraded. Besides, there may be some foreign words, letters, and abbreviations or numbers, mathematical expressions, unit symbols and so on. The above aspects become the main motivation to develop a TTS system for a special professional domain.Spaceflight is a very complex man-made artifact and works under rigorous condition with ultra high costliness. Therefore it is very important to carry out carefully testing on the systems of spaceflight during building or before launch. To achieve a very high reliability, the testing is very strict and even fussy in some degree. We introduce the TTS system to the testing procedure to improve the efficiency of testing, decrease the hardness of labor, enhance the working efficiency and keep the correctness of testing. With TTS system, the voice interaction takes the place of observing of screen, and this makes the testing work much easier. Besides, a euphony voice may relax the testers.Based on the requirements of spaceflight testing, a professional TTS system is built. This paper presents all the detail about the design and implementation of the system and the features of the TTS system can be concluded as follows:1. Application of TTS to Spacecreaft Testing system. Integrate the TTS into the testing subsystem and a basic human-machine interface, i.e. VoiceUI is formed.2. Basic speech unit selection, the system adopts various length unit settings. Chinese word, character and single-syllable are chosen as basic unit for mandarin. As for the words, abbreviations and characters in western language, the words and single/multi-syllable are chosen as basic unit. Such choice of speech unit may make a balance between the synthesis quality and size of unit library.3. Prosodic model, the parametric Target prosodic model is adopted in the system. Such model selection is due to the simple prosodic structure of the TTS system, in which most of the sentence to be synthesized is imperative or presentation. The mixed lingual sentences are always in a format that there are several western words, characters or phrase in a mandarin sentence, so the effect from western words to mandarin sentence is rather small.4. Corpus text collection, besides the ordinary text selection, more attention is paid to the collection of professional documents of spaceflight testing and other domain related. Some efforts are expended in building lexicon which should include key-terms in the domain of spaceflight and relates.5. Design of recording script, a strategy which can adjust the evaluation of candidate sentences dynamically is adopted to improve greedy algorithm. The result script yielded from such a strategy may share a similar word-frequency distribution as that of original text collection or have an arbitrary distribution so as to compensate some shorting of original text.6. Automatic unit pre-selection, some un-suitable speech unit can be found out by clustering and Artificial Neural Network (ANN) automatically.The contents of this paper come from the procedure of design and implementation of the TTS. The TTS system discussed in this paper has been applied in the testing system of satellites and good performance yielded.
Keywords/Search Tags:Speech Synthesis, Text to Speech, Multi-language Synthesis, Spaceflight Testing
PDF Full Text Request
Related items