Title :
Recognition Quality Improvement in Automatic Speech Recognition System for Polish
Author :
Wydra, Sebastian
Author_Institution :
Warsaw Univ. of Technol., Warsaw
Abstract :
This paper discusses a mixed parameterization modeling method in a speaker independent Automatic Speech Recognition (ASR) system for Polish keywords. The goal of this project is to improve the recognition quality in Polish commands and instructions recognition system based on mel-frequency cepstrum coefficients (MFCC) modeling. In this project we studied the recognition quality using a mixed speech modeling consisting of MFCC and prosodic parameters. We experimentally showed that the mixed modeling yields better recognition quality than a simple MFCC modeling.
Keywords :
cepstral analysis; hidden Markov models; speech recognition; Polish commands; Polish keywords; hidden Markov models; instructions recognition system; mel-frequency cepstrum coefficients modeling; mixed parameterization modeling method; mixed speech modeling; prosodic parameters; recognition quality improvement; speaker independent automatic speech recognition system; Automatic speech recognition; Cepstrum; Hidden Markov models; Markov processes; Mathematical model; Mel frequency cepstral coefficient; Random sequences; Robotics and automation; Speech recognition; Vocabulary; Hidden Markov Models; mel-frequency cepstrum coefficients; prosodic parameters; speech recognition;
Conference_Titel :
EUROCON, 2007. The International Conference on "Computer as a Tool"
Conference_Location :
Warsaw
Print_ISBN :
978-1-4244-0813-9
Electronic_ISBN :
978-1-4244-0813-9
DOI :
10.1109/EURCON.2007.4400509
