Speech Communication - Special issue on speech processing in adverse conditions
Hi-index | 0.00 |
In this paper, cepstral features derived from the Differentiated Relative Higher Order Autocorrelation Sequence Spectrum (DRHOASS) are proposed for improving the robustness of a speech recognizer in the presence of background noise. Proposed method is analyzed and compared in terms of the autocorrelation coefficients they employ with the traditional feature extraction methods based on Linear Pediction (LP) analysis. LP- based techniques used are Linear Predictive Cepstral Coefficients (LPCC), Short-Time Modified Coherence (SMC) and the One-Sided Autocorrelation Linear Prediction Coefficient (OSALPC). We evaluate the speech recognition performance of the proposed features on the Hindi isolated-word task and show that the proposed features show better recognition performance than the features derived from the robust liner prediction based methods for noisy speech.