Speech enhancement applied to speech recognition in noisy environments

Speech recognition systems work in practical environments using parameters training in quiet environments. The system performance is satisfactory when the environment is also quiet, but degrades quickly in noisy environments. This paper presents a noisy speech recognition system with added background noise so that the recognition system can work well both in quiet and noisy environments. Speech enhancement as the front-end processing module is used to improve the Signal-to-Noise Ratio (SNR) of the input signal for the recognition in the latter stages. Three different speech enhancement algorithms were tested with six types of noise. Experimental results show that the cascading of the speech enhancer and a Hidden Markov Model (HMM) based speech recognizer can significantly improve recognition accuracy in noisy environments without performance degradation for clean speech.