Method and decoder for detecting bad speech frames based on the spectral difference to the last uncorrupted frame

摘要

The present invention relates to a method for detecting bad speech frames e.g in mobile systems such as TETRA and a corresponding decoder.A spectral discontinuity between consecutive decoded frames is detected on the basis of line spectral pair (LSP) coefficients calculated from frame data (C). A change in gain between consecutive frames is also detected by extraction of frame gain parameter (E). Speech frames are considered errorfree if either of the spectral discontinuity or the change in gain lie below predetermined thresholds (D,F). In case both the spectral discontinuity and the change in gain exceed these thresholds, an energy difference is calculated to consecutive frames detecting an error only if a given threshold is exceeded (I). This way, large spectral discontinuities which occur often for low energy levels do not automatically lead to the false detection of frame errors.