Sound Structure Description for speech recognition






< Sound Structure Description >





No. 8,  7  February  2009

previous pages

Complete sound structure description  is that describe history from the origin of the sound,  all transformation process to the finish of the sound, including sound effects towards our ears. But, this history description method is not so useful, because  in actual fact  it's hard to get whole history.  Other useful alternative method development is necessary.  An idea of sound structure description method  is that  decompose a sound  into combinations of  characteristic features which have relation to each other. Past transformation process reflects relation. Characteristic features consist of  two or more different kinds of type.

For example,  when mouth works as like a resonance tube, each relative waves are generated  simultaneously.  Analysis how to relate each waves. Using some generator model which simulates, by  mathematical optimize method    fitting to the generator model , and  statistics analysis about model's variables in order to make use of pattern recognition method on probability. Due to movement limitation of mouth, some pole points are  brought about  and  they may be center points in statistics.

Maybe, methods to detect  these characteristics based on non-linear, should be discovered by human, not by machine. Once, these methods are well understood,  then  they will be able to be transformed  systematic method, that is  like logical synthesis method.






No. 8,  7  December  2008


This page first established on 6 September 2008.