Research And Improvement Of Speech Enhancement Algorithms Base On Single-Channel

There is a variety of noise interference in the speech communication, and serious noise interference may lead to auditory fatigue or performance deterioration of speech processing system. In order to recover the original speech signal as much as possible from the speech signal containing all kinds of noise and improve the quality and intelligibility of speech signal, we need to use speech enhancement technology to suppress or reduce the noise. Speech enhancement can be divided into single-channel and double-channel and multi-channel according to different ways of signal acquisition. At present most of signal processing systems are based on single-channel due to less access to the information, compared to double-channel and multi-channel. As a result, it is very important to study and improve speech quality based on single-channel. Therefore,this paper consists of the following work:1. Different noises have different characteristics, so in order to achieve the best effect of speech enhancement in practice, we must choose different speech enhancement algorithms based on different noises. For this reason, we in this paper deeply studied the spectrum subtraction, wiener filtering algorithm and minimum mean square error algorithm, and a large number of tests under Gaussian white noise, pink noise and babble noise were done. The results show that the three algorithms under different noises can improve speech quality, but not always improve speech intelligibility.2. Estimate of the noise in the speech enhancement technology is very important, and inaccurate estimation can result in speech distortion. For example, low estimation will lead to big background noise, and over-estimation will weaken the faint voice messages. In this paper, VAD algorithm and average time recursive algorithm based on a posteriori were studied. Due to the step phenomenon of smoothing factor (0 or 1), an improved measure was proposed to ensure the smoothing factor in a quite reasonable range. The measured results show that the improved algorithm in low SNR environment has a good function.3. Spectral subtraction algorithm has simple, efficient and high real-time performance with a broad applicable scope, but output of music noise is a disadvantage. In order to solve the music noise, people come up with a lot of improvement measures. To reduce the effect of music noise, one or more correction coefficients are used commonly in the process of subtraction, but the adaptability is very bad because it is based on experiment. Therefore, the spectral subtraction based on optimal control of parameters was studied in the paper, and optimal correction coefficient was determined by estimating a prior with an improved decision method. The results showed that the increase of speech intelligibility is limited, but speech quality is improved.