A low delay, variable resolution, perfect reconstruction spectral analysis-synthesis system for speech enhancement

The choice of the window function and window length in short time analysis-synthesis (AS) systems based on the discrete Fourier transform (DFT) has to balance conflicting requirements: Long windows provide high spectral resolution while short windows allow for high temporal resolution. Furthermore, for many applications a low algorithmic delay is desirable. Therefore, long standard windows such as the Hann or Hamming windows cannot be used. In this contribution we propose a novel AS system which achieves perfect reconstruction (PR) and a low delay by using a variable length analysis window and a relatively short synthesis window. The variable length analysis windows allow a spectral analysis that is adapted to the signals span of stationarity. The AS windows are designed such that they can be switched at any time instant without violating PR. We show that the spectral representation of typical speech data is improved as compared to AS systems with standard windows.