Application of Kullback -Leibler Metric to Speech Recognition

Title: Application of Kullback -Leibler Metric to Speech Recognition

Research Question: Can the Kullback -Leibler divergence be applied to speech recognition, and if so, how effective are the resulting algorithms?

Methodology: The study proposes three algorithms for speech recognition: correlation, spectral, and filter algorithms. These algorithms implement the Kullback -Leibler divergence criterion to analyze the information characteristics of input signals. The algorithms are designed to handle variable signal duration, using a computationally efficient scheme of signal alignment. The effectiveness of these algorithms is compared to other existing methods.

Results: The study found that the Kullback -Leibler divergence criterion can be successfully applied to speech recognition. The correlation algorithm, spectral algorithm, and filter algorithm showed promising results in terms of performance and efficiency. The study also provided recommendations for choosing appropriate model parameters and optimizing the algorithms.

Implications: The application of Kullback -Leibler divergence to speech recognition has significant implications for the field. It offers a new approach to developing more robust and effective speech recognition algorithms, which could lead to the widespread implementation of speech interfaces in various automatic systems. The study's findings also contribute to the ongoing research in signal processing and machine learning.

Link to Article: https://arxiv.org/abs/0310023v1 Authors: arXiv ID: 0310023v1