Y.-W. Bai and J.-L. Chen (Taiwan)
Majority Voting Mechanism, Speech RecognitionProbability
To improve the accuracy of speech recognition, some researchers have proposed the integration method of majority voting through multiple recognition modules. In this paper, we propose a mathematically proof and present experimental result of combining the individual modules into a majority voting mechanism. The mathematical proof is based on the summation of the recognition probability of the each enumeration. By using the typical theory of limitation, we obtain the improvement of the overall recognition probability when the number of modules of the speech recognition increase and convergence will exist. From the comparison of the probability equations, we learn the overall recognition probability will increase if the recognition probability of the single module is greater than 0.5. The experiments for the speech recognition by a majority voting design also show the increase in overall recognition probability through the integrated system.
Important Links:
Go Back
