Speech enhancement algorithms for multimedia applications

点击翻译按钮获取中文摘要

Speech enhancement plays a fundamental role in many applications such as teleconferencing systems, speech recognizers, human-machine interfaces, hands-free communications, mobile phones, and hearing aids. Indeed, the noise is around us and everywhere. So when a signal of interest (usually speech) is picked up by a microphone, it is always contaminated by noise, reverberation, and other undesired signals. The objective is then to clean up the noisy signal with digital signal processing tools without damaging much the desired speech signal. The problem of speech enhancement is an old one and has been around for more than 40 years. This should not come as a surprise since, contrary to what some people may believe, it is a very difficult problem for mainly two reasons. First, the nature and characteristics of the noise signals can change dramatically in time and application to application. It is therefore laborious to find versatile algorithms that really work in different practical environments. Second, the performance measure can also be defined differently for each application. Two perceptual criteria are widely used to measure the performance: quality and intelligibility. While the former is subjective (it reflects individual preferences of listeners), the latter is objective (it gives the percentage of words that could be correctly identified by listeners). It is very hard to satisfy both at the same time. So this research proposal aims at developing the next-generation speech enhancement algorithms with one and, especially, multiple microphones (or microphone arrays) with the clear objective to reduce the noise and other interference signals as much as possible with little (or no) distortion of the desired speech signal. In other words, we will try to improve the quality as much as possible without affecting the intelligibility. We will also try to develop practical algorithms that can be easy to use and easy to tune in the real world and for most applications. To achieve this goal, we will rely on some strong ideas and results we have developed recently.