Audio Samples for "SoundSpring: Loss-Resilient Audio Transceiver with Dual-Functional Masked Language Modeling"

Neural audio codec:


SoundStream: Neural audio codec using residual vector quantization (RVQ) by Google Research. SoundSpring employs the same neural audio encoder and decoder.
N. Zeghidour, et.al ``Soundstream: An end-to-end neural audio codec,'' IEEE/ACM Trans. Audio Speech Language Process., vol. 30, pp. 495–507, 2021.

Traditional audio codec:


AMR-WB: B. Bessette, et.al ``The adaptive multirate wideband speech codec (AMR-WB),'' IEEE Trans. Speech Audio Process., 2002
Opus: Low bit-rate redundancy (LBRR) is available in 18 kbps.
J.-M. Valin, et.al, ``Definition of the Opus audio codec,'' IETF, 2012.
AAC: Advanced audio coding, ISO/IEC 13818-7:1997

Speech, 16 kHz Mono

10 % Random loss

30 % Random loss

WLAN Trace

Original

Original

Original

AMR-WB 9kbps

AMR-WB 9kbps

AMR-WB 9kbps

Opus @18kbps (LBRR)

Opus @18kbps (LBRR)

Opus @18kbps (LBRR)

Opus @9kbps

Opus @9kbps

Opus @9kbps

SoundSpring + FEC @5.7kbps

SoundSpring + FEC @5.7kbps

SoundSpring + FEC @12kbps

SoundSpring @9.3kbps

SoundSpring + FEC @12kbps

SoundStream + FEC @14.8kbps

SoundStream + FD-PLC @6kbps

SoundStream + FD-PLC @6kbps

SoundStream + FD-PLC @12kbps

Music, 48 kHz Stereo

10 % Random loss

30 % Random loss

WLAN Trace

Original audio

Opus @18kbps

AAC @32kbps

SoundSpring @9.2kbps

SoundSpring + FEC @12.1kbps

SoundStream + FEC @14.8kbps

All rights reserved