List your company

Neural Speech Coding on LibriSpeech clean (test)

4.64PESQGround Truth
1.11442.02972.9453.8603May 19, 2026May 21, 2026May 24, 2026May 27, 2026May 29, 2026Jun 1, 2026Jun 4, 2026

Evaluation results

MethodLinks
Ground Truth# Param=-, # hours=-, frame rate=-, bps=-2026.054.641004.091001.5—————
DualCodecTraining updates=300k2026.062.954—3.950.823—12.757.494.613.933.87
HC-SED-AEDTraining updates=300k2026.062.894—3.990.816—12.967.324.43.833.63
DAC (Distill)Training updates=300k, Process=Distill2026.062.834—3.9260.776—18.0110.555.344.824.14
DualCodec (open-source)Training updates=300k, Source=open-source2026.062.833—3.7710.92—17.2314.048.225.026.32
Mimi (open-source)Training updates=300k, Source=open-source2026.062.589—3.77450.8374—35.778.586.324.154.14
FlexiCodec# Param=450M, # hours=54k, frame rate=6.25, bps=6402026.052.2904.15712.57—————
SAC# Param=533M, # hours=20k, frame rate=12.5/25, bps=5252026.052.16904.27782—————
StableCodec-700# Param=950M, # hours=105k, frame rate=25, bps=7002026.051.92894.31583.91—————
StableCodec-400# Param=950M, # hours=105k, frame rate=25, bps=4002026.051.92894.31534.88—————
ClariCodec (w/o RL)# Param=301M, # hours=50k, frame rate=12.5, bps=3002026.051.88874.12504.64—————
ClariCodec# Param=301M, # hours=50k, frame rate=12.5, bps=3002026.051.87874.16503.55—————
WavTokenizer# Param=72M, # hours=8k, frame rate=40, bps=4802026.051.63853.57517.38—————
SemantiCodec# Param=507M, # hours=37.6k, frame rate=12.5/12.5, bps=312.52026.051.38782.723422.7—————
SoCodec# Param=54M, # hours=7.2k, frame rate=8.3, bps=4662026.051.2892.5395.59—————
Encodec# Param=15M, # hours=17.5k, frame rate=10, bps=7502026.051.25771.252516.1—————