Signal modification for coding purely voiced sections in a wideband ACELP speech coder

In narrowband CELP coding, signal modification is often employed to improve pitch prediction at lowest bit rates. In this paper, we extend signal modification for wideband speech coding. A method for identifying, modifying, and coding purely voiced speech frames in a variable bit rate wideband ACELP coder is presented. Signal modification enables coding of purely voiced frames with a low bit rate mode. Modification is performed frame and pitch synchronously preserving the original time scale at the end of each frame. Since no modification is allowed at the frame end, the delay parameter of pitch prediction is prone to oscillations. These oscillations can be reduced by choosing a proper delay contour for interpolating the parameter over the modified frame We also demonstrate that the signal modification algorithm can be employed for detecting reliably frames suitable for signal modification. Informal listening tests indicate good performance for the proposed method.