A Hybrid Structure Speech Coding Scheme Based on MELPe and LPCNet

Although the conventional speech coding algorithms can achieve 1200bps, 600bps or even lower, the synthesized speech quality becomes worse and worse, and the improvement effect is limited. Several speech synthesis models based on deep learning proposed in recent years make it possible to synthesize high-quality speech. This paper integrates LPCNet into the traditional MELPe algorithm to finish the speech synthesis task. Experiments demonstrate that the quality of this scheme is similar to that of MELPe at 1.07kbps, which can be used as a new idea to study the ultra-low rate MELPe.