Audio samples for HILCodec

[github] [paper]

Sampling rate: 24khz, Mono channel

2 versions of HILCodec are available: One trained on general audio including clean speech, noisy speech, and music. The other trained on clean speech only.

mushra scores

Models trained on general audio

Low bitrates

Reference HILCodec (3kbps) EnCodec (3kbps) Descript (3kbps) EVS (5.9kbps)
clean speech
noisy speech
music

High bitrates

Reference HILCodec (9kbps) EnCodec (9kbps) Descript (9kbps) EVS (16.4kbps)
clean speech
noisy speech
music

Models trained on clean speech

Low bitrates

Reference HILCodec (3kbps) HiFiCodec (3kbps) AudioDec (6.4kbps) EVS (5.9kbps)
clean speech 1
clean speech 2
clean speech 3

>