Speech Coding on Emilia clean subset (test)
3.646DNSMOSRFSQ-4S-NU-LN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RFSQ-4S-NU-LNBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=4, Bit Allocation=Non-Uniform (8, 6, 5, 5), Conditioning Strategy=LayerNorm2025.08 | 3.646 | 3.8 | |
| RFSQ-4S-Uni-LNBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=4, Bit Allocation=Uniform, Conditioning Strategy=LayerNorm2025.08 | 3.598 | 3.83 | |
| RVQ-4×64Bitrate=1.8 kbps, Sampling Rate=24 kHz, Stages=4, Quantization Architecture=RVQ, Codebook size=642025.08 | 3.518 | 3.81 | |
| RFSQ-4S-NU-ScaleBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=4, Bit Allocation=Non-Uniform, Conditioning Strategy=Scale2025.08 | 3.421 | 3.7 | |
| RFSQ-8S-Uni-LNBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=8, Bit Allocation=Uniform, Conditioning Strategy=LayerNorm2025.08 | 3.356 | 3.65 | |
| RFSQ-2S-NU-LNBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=2, Bit Allocation=Non-Uniform, Conditioning Strategy=LayerNorm2025.08 | 3.289 | 3.62 | |
| RFSQ-4S-NU-NoBitrate=1.8 kbps, Sampling Rate=24 kHz, Number of Stages=4, Bit Allocation=Non-Uniform, Conditioning Strategy=none2025.08 | 3.187 | 3.58 | |
| FSQ-4D-UniformBitrate=1.8 kbps, Sampling Rate=24 kHz, Quantization Architecture=Single-stage FSQ, Bit Allocation=Uniform2025.08 | 2.965 | 3.51 | |
| LFQ-24DBitrate=1.8 kbps, Sampling Rate=24 kHz, Quantization Architecture=Lookup-Free Quantization2025.08 | 2.814 | 3.52 | |
| VQ-EMA-4×64-PQBitrate=1.8 kbps, Sampling Rate=24 kHz, Stages=4, Quantization Architecture=VQ-EMA, Type=Product Quantization2025.08 | 2.687 | 3.42 |