A Closer Look at Neural Codec Resynthesis:
Bridging the Gap between Codec and Waveform Generation

Anonymous submission to Interspeech 2024

Codec Resynthesis Results

see paper for detailed explanation

Codec resynthesis results, see paper for detailed explanation. Number of function evaluations (NFE) reflects the number of forward passes required for synthesis, i.e., the inference cost.

Audio Samples

Demo samples are used in the MOS test.
!!! Wearing headphones is strongly recommended to judge the audio quality !!!
                             
Input
(1st RVQ code)
Coarse-to-fine
(NFE=7)
One step
(NFE=1)
Schrödinger Bridge
(NFE=1)
Schrödinger Bridge
(NFE=7)
Schrödinger Bridge
(NFE=16)
Reference