Products of the same type
Products of the same team

Pronunciation Model

Pronunciation variation modelling system

First version: March 1999 Last version: December 1999
Application:  
  Can be built into a system for speech synthesis and into a system of automatic speech recognition.  
Description:  
  The system inputs a text in standard spelling, and outputs a transcription of its reading in accordance with a specific regional dialect. The system produces Russian connected speech. The user can independently tune the program for simulation of the required regional pronunciation style.
The system includes 2 main modules: an Adaptive Automatic Transcriber (AAT) and a corpus of pronunciation configurations.
AAT provides the possibility for generating different pronunciation forms on the basis of any orthographic form by means of including/excluding and modifying normative phonological and phonetic rules in accordance with the following phonetic conditions: context influence, the quality of the stressed, pre-stressed and post-stressed vowels in the word, word and syntagmatic stress position. There is also a possibility for generating pronunciaton variants in a concrete morpheme or a lexeme.
The corpus of pronunciation configurations includes various sets of phonological and phonetic rules realised in different conditions. Currently there is a corpus of several pronunciation configurations for representation of the following types of pronunciation variations: 		 
 
  1. orthoepic variation within the limits of the literary pronunciation standard (careful speech);
  2. pronunciation variation in casual speech (certain sound variations are taken into account, e.g., the increase of coarticulatory and assimilation effects, elisions etc.)
  3. regional variation (the influence of several main Russian dialects is taken into account).
 
  The system uses an auxilliary module (UDAR 7.5), developed by I.V. Zharkov (see the description).
This module includes a vocabulary containing about 100,000 lexemes with the indication of the main morphological classes and a stress intonational transcriber of Russian texts.		  
Languages processed:  
  Russian  
Type of processing:  
  Stress Intonational Transcriptor;
Vocabulary of 100 000 lexems.
Application of Adaptive Automatic Transcriber, basing on the Corpus of pronunciation configurations		  
Hardware/software requirements:
  IBM PC compatible (Pentium II or higher), Windows 95/98, Object Pascal (Delphi)  
Distribution: Prototype  
Developer team: Contact person:  
Department for phonetics and methods of teaching foreign languages, SPbGU Pavel A. Skrelin
Phone: (+7 812) 328 95 65
E-mail: paul@PS1089.spb.edu, paul@phonet.lang.pu.ru