Research On Speech Quality Evaluation Algorithm Based On Fusion Technology

Speech evaluation has important application value in the modern informationsociety. Currently, most speech evaluation systems at home and abroad are based onspeech recognition. However, the relevance between speech evaluation which arebased on speech recognition technology and subjective evaluation of experts isrelatively low. Therefore, the overall performance of the system can not providesatisfying result.This study combines with the phonetics, linguistics and other disciplinesrelevant theory and knowledge. It makes an objective evaluation to speech qualityfrom the aspects of acoustics, rhythm, speech perception, and establishes the speechquality evaluation system on the basis of multiple fusion models. The system designis divided into two parts which are theory researchs and the software simulation.Thispaper focuses on the research of algorithm of acoustic models, perception modelsand rhythm models. It objectively appraises the quality of speech from differentangles. The passage mainly researchs the problem of data fuse between differentmachine scores. The influence on the performance of the system which is caused bythe mapping function and data fusion methods is analyzed. In order to validate thefeasibility of this scheme, this paper builds up a speech quality evaluation system byMATLAB, Visual Studio2008. This thesis achieves the software simulation and theinterface design by setting up a small speech database. It completes the scoringalgorithm simulation, the pitch detection method simulation and the extractingcharacteristic parameters simulation through MATLAB. It accomplishes thepermissions landing interface, the inputing speech signal, menu interface and so onby Visual Studio2008.Experimental results show that the pronunciation evaluation method cansignificantly improve the correlation between the objective evaluation and subjectiveevaluation. This system is valuable in advanced research and development.