On the use of RSA and DRA to improve the robustness of continuous speech recognition systems in adverse conditions

This paper describes a noise-robust feature extraction technique for continuous speech recognition. The proposed method can thus extract speech features in adverse conditions. Our method combines both running spectrum analysis (RSA) and dynamic range adjustment (DRA) and focuses on speech feature adjustment. DRA is deployed after RSA in order to improve speech recognition performance. We use robust speech features to build a continuous-speech recognition model whose effectiveness is proven. Simulation results are provided for 10 dB and 15 dB noise conditions. In various noise conditions, the average recognition rate improves by more than 10% at 10 dB SNR.