Speech Enhancement in No Environme

This paper presents a speech enhancement using a noise estimation based on the ratio of the noisy speech and its minimum (NSMR) for non-stationary noise environments. The noise estimator is a very simple but highly effective real time approach for single channel noise reduction. The enhanced speech is free of musical tones and reverberation artifacts and sounds very natural compared to methods using other short-time spectrum attenuation techniques. The performance is measured by the segmental signal-to-noise ratio and MOS tests. To judge the performance the recognition accuracy of an Automatic Speech Recognition (ASR) system using Mel-scale Frequency Cepstral Coefficients (MFCC) features is measured with and without noise reduction. In another experiment we apply the NSMR noise reduction method to speech reconstruction at the back-end of a distributed speech recognition (DSR) system under various noise conditions.