An Efficient Block-based Dynamic Range Adjustment Method in Noise-robust Continuous Speech Recognition

This paper proposes a new technique for speech feature estimation under noise circumstances. This new approach yields noise-robust continuous speech recognition (CSR). Noiserobust techniques for isolated word speech recognition typically employ the running spectrum analysis (RSA), the running spectrum filtering (RSF) and the dynamic range adjustment (DRA) methods. Among them, only RSA has been applied into a CSR system. However, we propose an enhanced DRA for a noise-robust CSR system. Thus, in the speech recognition stage, the continuous speech waveform is automatically divided into short blocks and DRA is applied to these blocks. We find that the proposed method improves recognition performance under several different noise and SNR conditions.