Results

**1 - 2**of**2**### THE STOCHASTIC WEIGHTED VITERBI ALGORITHM: A FRAME WORK TO COMPENSATE ADDITIVE NOISE AND LOW – BIT RATE CODING DISTORTION

"... A solution to the problem of speech recognition with signals corrupted by additive noise and distorted by low-bit rate coders is presented in this paper. The additive noise and the coding distortion are cancelled according to the following scheme: firstly, the pdf of the clean coded-decoced speech i ..."

Abstract
- Add to MetaCart

(Show Context)
A solution to the problem of speech recognition with signals corrupted by additive noise and distorted by low-bit rate coders is presented in this paper. The additive noise and the coding distortion are cancelled according to the following scheme: firstly, the pdf of the clean coded-decoced speech is estimated with an additive noise model; second, the pdf of the clean uncoded signal is also estimated with a coding distortion model; and finally, the HMM is compensated by using the expected value of the observation pdf in the context of the stochastic weighted Viterbi (SWV) algorithm. The approach leads to reductions as high as 50 % or 60 % in word error rate. 1.

### FEATURE-DEPENDENT COMPENSATION IN SPEECH RECOGNITION

"... Several mismatch conditions can be modeled as an additive bias. This bias is considered independent of the observation vectors, although this approximation is not always accurate. In this paper the dependence of the bias on the observation vectors is taken into consideration in the context of compen ..."

Abstract
- Add to MetaCart

(Show Context)
Several mismatch conditions can be modeled as an additive bias. This bias is considered independent of the observation vectors, although this approximation is not always accurate. In this paper the dependence of the bias on the observation vectors is taken into consideration in the context of compensating the GSM coding distortion in speech recognition. However, the results presented here can easily be generalized to deal with other types of mismatch. The coding-decoding distortion is modeled here as feature-dependent. This model is employed to propose an Expectation-Maximization (EM) estimation algorithm of the codingdecoding distortion that is able to cancel the effect of GSM coder with as few as one adapting utterance. Finally, the feature-dependent adaptation can give word error rate (WER) 26 % lower than the featureindependent model. 1.