SoundStream: An End-to-End Neural Audio Codec

TL;DR

SoundStream is a neural audio codec that efficiently compresses various audio types at low bitrates, supporting real-time streaming and noise suppression, with high quality comparable to or better than existing codecs.

Contribution

Introduces SoundStream, a neural audio codec capable of variable bitrate compression, real-time streaming, and joint enhancement, trained with adversarial and reconstruction losses for high audio quality.

Findings

Outperforms Opus at 12kbps and approaches EVS at 9.6kbps in subjective quality.

Supports variable bitrates from 3kbps to 18kbps with negligible quality loss.

Enables joint compression and noise suppression without additional latency.

Abstract

We present SoundStream, a novel neural audio codec that can efficiently compress speech, music and general audio at bitrates normally targeted by speech-tailored codecs. SoundStream relies on a model architecture composed by a fully convolutional encoder/decoder network and a residual vector quantizer, which are trained jointly end-to-end. Training leverages recent advances in text-to-speech and speech enhancement, which combine adversarial and reconstruction losses to allow the generation of high-quality audio content from quantized embeddings. By training with structured dropout applied to quantizer layers, a single model can operate across variable bitrates from 3kbps to 18kbps, with a negligible quality loss when compared with models trained at fixed bitrates. In addition, the model is amenable to a low latency implementation, which supports streamable inference and runs in real time on a smartphone CPU. In subjective evaluations using audio at 24kHz sampling rate, SoundStream at 3kbps outperforms Opus at 12kbps and approaches EVS at 9.6kbps. Moreover, we are able to perform joint compression and enhancement either at the encoder or at the decoder side with no additional latency, which we demonstrate through background noise suppression for speech.