New ideas for waveform interpolation

Waveform interpolation speech compression has been shown to provide high quality speech at very low bit rates, but has not yet been extensively deployed due to extremely high computational complexity and memory requirements. We present two novel techniques that ameliorate the situation. First, we propose a new efficient algorithm for characteristic waveform (CW) alignment. Second, we augment the standard linear-filtering decomposition of the CW into Slowly Evolving Waveform (SEW) and Rapidly Evolving Waveform (REW) by a voicing-dependent factor. These techniques lead to reduction of memory requirements and computation without significant degradation of the speech quality. As an added benefit these techniques also remove an artifact typically associated with waveform interpolation encoders.