Real-Time Packet Loss Concealment With Mixed Generative and Predictive Model

TL;DR

This paper introduces a hybrid neural network for real-time packet loss concealment in speech communication, combining generative and predictive models to produce natural-sounding speech recovery that outperforms traditional methods.

Contribution

A novel hybrid neural PLC architecture that integrates generative and predictive models, achieving superior real-time speech concealment quality and versatility across codecs.

Findings

Outperforms conventional PLC algorithms in naturalness and quality.

Ranked second in the Interspeech 2022 PLC Challenge.

Applicable to both uncompressed audio and modern speech codecs.

Abstract

As deep speech enhancement algorithms have recently demonstrated capabilities greatly surpassing their traditional counterparts for suppressing noise, reverberation and echo, attention is turning to the problem of packet loss concealment (PLC). PLC is a challenging task because it not only involves real-time speech synthesis, but also frequent transitions between the received audio and the synthesized concealment. We propose a hybrid neural PLC architecture where the missing speech is synthesized using a generative model conditioned using a predictive model. The resulting algorithm achieves natural concealment that surpasses the quality of existing conventional PLC algorithms and ranked second in the Interspeech 2022 PLC Challenge. We show that our solution not only works for uncompressed audio, but is also applicable to a modern speech codec.