Using adaptive signal limiter together with weighting techniques for noisy speech recognition

In an automatic speech recognition (ASR) system, environmental mismatch between speech models and speech test utterances causes serious performance degradation in spectral analysis methods. To alleviate this environmental mismatch problem, smoothing processes and weighting techniques are two of the most widely used methods. An investigation is presented into the feasibility of combining an adaptive signal limiter with a weighting technique for seeking further performance improvement in noisy speech recognition.