Enhancement and Bandwidth Compression of Noisy Speech by Estimation of Speech and Its Model Parameters
Abstract
The problem of enhancement and bandwidth compression of noisy speech is formulated as a parameter estimation problem, in which speech and its model parameters are estimated from the noisy speech based on the MAP estimation procedure. Such an approach leads to two algorithms which require solving sets of linear equations in an iterative manner. Some approximations of the two algorithms lead to two systems which are computationally simpler than the two algorithms by taking advantage of a high speed FFT algorithm. As a preliminary investigation into the performance of the class of systems developed, two systems are implemented and applied to both real and synthetic speech data. An objective and informal subjective evaluation indicate that the systems implemented perform well as enhancement and potential bandwidth compression systems of noisy speech.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1978
- Accession Number
- ADA081948
Entities
People
- Jae S. Lim
Organizations
- Massachusetts Institute of Technology