Wavenet based low rate speech coding

TL;DR

This paper demonstrates that a WaveNet generative model can significantly improve low-rate speech coding quality, enabling high-fidelity speech synthesis from minimal bit streams, with additional benefits like bandwidth extension and speaker recognition preservation.

Contribution

It introduces a WaveNet-based approach for low-rate speech coding that outperforms traditional parametric coders and compares favorably with waveform coders, showing high quality and robustness.

Findings

High-quality speech synthesis at 2.4 kb/s

WaveNet-based coder outperforms traditional parametric coders

Produced speech maintains speaker recognition and bandwidth extension

Abstract

Traditional parametric coding of speech facilitates low rate but provides poor reconstruction quality because of the inadequacy of the model used. We describe how a WaveNet generative speech model can be used to generate high quality speech from the bit stream of a standard parametric coder operating at 2.4 kb/s. We compare this parametric coder with a waveform coder based on the same generative model and show that approximating the signal waveform incurs a large rate penalty. Our experiments confirm the high performance of the WaveNet based coder and show that the speech produced by the system is able to additionally perform implicit bandwidth extension and does not significantly impair recognition of the original speaker for the human listener, even when that speaker has not been used during the training of the generative model.