/pdf/coarticulation-knowledge-for-a-text-to-speech-system-for-an-45ixrh3a6v.pdf

Coarticulation knowledge for a text-to-speech system for an indian language

This thesis is an investigation into the ability of artificial neural networks to learn to map from a symbolic representation of CVC triphones to a continuous repre¬ sentation of vowel formant tracks, and the influence of a number of factors on that ability. This mapping is interesting because, apart from being a necessary part of any text to speech system and not having any accepted definitive solu¬ tion, it is from a discrete symbolic representation to a continuous non-symbolic representation. Neural networks provide one method of automatically learning such mappings and prove to be capable of doing so in this particular case. The input representation used appears to have little effect on the perfor¬ mance of the neural networks. A feature based representation does no better than a 1-of-n coding of the phonemes. The representation of the vowel formant tracks, produced as output of the neural networks, has a far greater effect on performance. Simple representations consisting of the initial, central and final frequencies of the formant tracks out-perform polynomial and Fourier coefficient representations which encode more information about the shape of the formant tracks. The back-propagation and conjugate gradient neural network training algo¬ rithms produced neural networks with similar performance, and the use of crossvalidation made no difference in generalisation (although the cross-validation data set was far too small). Interestingly, neural networks with no hidden layer proved to be as capable of learning the mapping as those with a hidden layer. I have also derived a relationship predicting the result of a modified rhyme

Vowel synthesis using feed-forward neural networks

This paper describes a highway advisory information radio system (highway radio system) and a speech synthesizer that is used in it. The Japan Highway Public Corporation has been developing highway radio systems which collect traffic information and automatically broadcast it using a speech synthesizer in order to improve safety and convenience of driving. In such systems, detailed and appropriate information, immediacy, and plain style message are required. However, this is very difficult and an usual speech synthesizer to handle a large variety of messages is required. The authors have developed a new speech synthesis method. In this method, pitch periods of an utterance are modified and sentence intonation is generated automatically when speech data are concatenated. Thus natural sounding sentence speech can be synthesized from shorter speech units such as words and phrases. Moreover, the highway radio system updates messages every five minutes with high speed processing, and generates appropriate messages based on evaluation of the importance of collected information. This highway radio system has been working since November 1992 on the Tohoku Expressway and two other expressways. >

A speech synthesizer for a highway advisory information radio system

Neural network based concatenation method of synthesis units for synthesis by rule.

Neural network based spectral interpolation method for speech synthesis by rule.

A text‐to‐speech synthesis system for Japanese is described. In synthesis of Japanese, generation of fundamental frequency pattern is one of the most important problems. We proposed the prosody rules that generate intonation pattern of any Japanese text using syntactic information. The prosody rules consist of modification rules, syntax rules, and a generation rule. Connecting several words, the system makes an accent phrase and gives it an accent using the dictionary and the rules (modification rules). The parameters used in a subsequent generation rule are modified on the basis of the syntactic structure of the input text (syntax rules). Then, a fundamental frequency pattern is generated from these parameters (generation rule). In this system, the synthesizer parameter is Cepstrum. Parameter rules assign segment duration and concatenate CV syllables Cepstra. Thus, this system translates any text of Japanese written with the syntactic symbols into good quality and natural speech.

Yasushi Ishikawa

Papers

Neural network based concatenation method of synthesis units for synthesis by rule.

Neural network based spectral interpolation method for speech synthesis by rule.

Text‐to‐speech synthesis of Japanese based on the Cepstrum‐CV method