prism-v/open_vocoder_leaderboard
0
1model_id,model_name,architecture_family,track,sampling_rate_hz,params_m,overall_score,pesq,stoi,mcd_db,lsd_db,utmos,nisqa,delta_wer_pct,rtf,speedup_x,peak_vram_mb,edge_feasible,is_pareto,checkpoint_url,paper_url,license,author,year,rank2vocos_mel_24khz,Vocos,Fourier / iSTFT ConvNeXt,24kHz Primary Benchmark,24000,13.58,82.8,3.714,0.971,38.35,6.99,3.84,3.91,3.42,0.0027,370.4,327.6,Yes,True,https://huggingface.co/charactr/vocos-mel-24khz,https://arxiv.org/abs/2306.00814,MIT,Charactr Inc.,2023,13flow2gan,Flow2GAN (4-step),Flow Matching + GAN,24kHz Primary Benchmark,24000,79.08,81.5,4.469,0.994,29.36,6.46,4.05,4.1,2.95,0.0833,12.0,1037.3,Yes,True,https://huggingface.co/k2-fsa/Flow2GAN,https://arxiv.org/abs/2405.08819,Apache-2.0,K2-FSA,2024,24rndvoc,RNDVoC,Random Distortion GAN,24kHz Primary Benchmark,24000,3.94,80.7,4.188,0.985,38.32,6.54,3.88,3.92,3.35,0.0284,35.2,1925.9,Yes,True,https://huggingface.co/AndongLi/RNDVoC,https://arxiv.org/abs/2406.01257,MIT,CAS / Li et al.,2024,35bridgevoc,BridgeVoC,Brownian Bridge SDE,24kHz Primary Benchmark,24000,7.89,79.6,4.377,0.99,38.99,6.57,3.92,3.95,3.25,0.0839,11.9,591.4,Yes,False,https://huggingface.co/AndongLi/BridgeVoC,https://arxiv.org/abs/2406.01258,MIT,CAS / Li et al.,2024,46comvo_base,ComVo-Base,Complex-Valued Fourier,24kHz Primary Benchmark,24000,13.28,78.8,3.816,0.975,42.43,7.05,3.79,3.82,3.65,0.0123,81.3,643.4,Yes,False,https://huggingface.co/hsoh/ComVo-base,https://arxiv.org/abs/2406.19794,Apache-2.0,Seoul National Univ.,2024,57periodwave_turbo,PeriodWave-Turbo (4-step),Diffusion / Turbo SDE,24kHz Primary Benchmark,24000,29.81,78.6,4.369,0.988,38.95,6.93,4.08,4.12,2.88,0.148,6.8,2746.9,Yes,False,https://github.com/kaist-dsp/PeriodWave,https://arxiv.org/abs/2408.06945,MIT,KAIST,2024,68bigvgan_v2_24khz_100band_256x,BigVGAN-v2 (112M),Anti-Aliased Snake GAN,24kHz Primary Benchmark,24000,112.45,77.8,4.381,0.992,25.38,6.55,4.12,4.18,2.75,0.2208,4.5,2719.6,Yes,False,https://huggingface.co/nvidia/bigvgan_v2_24khz_100band_256x,https://arxiv.org/abs/2206.04658,MIT,NVIDIA,2024,79comvo_large,ComVo-Large,Complex-Valued Fourier,24kHz Primary Benchmark,24000,115.35,77.6,4.03,0.984,31.53,6.84,3.96,4.02,3.1,0.0525,19.0,1857.0,Yes,False,https://huggingface.co/hsoh/ComVo-large,https://arxiv.org/abs/2406.19794,Apache-2.0,Seoul National Univ.,2024,810griffin_lim,Griffin-Lim STFT,Algorithmic DSP Baseline,24kHz Primary Benchmark,24000,0.0,73.7,4.014,0.995,17.68,3.01,2.15,2.2,8.9,0.0057,175.4,686.9,Yes,True,N/A (Algorithmic),https://ieeexplore.ieee.org/document/1172092,Public Domain,Griffin & Lim,1984,911freev,FreeV,Pseudo-Inverse Mel GAN,22.05kHz Legacy Track,22050,18.22,73.5,3.149,0.948,68.08,6.98,3.52,3.58,4.2,0.0033,303.0,387.5,Yes,False,https://huggingface.co/Bakerbunker/FreeV_Model_Logs,https://arxiv.org/abs/2405.15842,Apache-2.0,Baker et al.,2024,1012bigvgan_base_24khz_100band,BigVGAN-Base (14M),Anti-Aliased Snake GAN,24kHz Primary Benchmark,24000,14.03,71.2,3.856,0.978,39.48,6.99,3.62,3.68,3.9,0.1103,9.1,3094.8,Yes,False,https://huggingface.co/nvidia/bigvgan_base_24khz_100band,https://arxiv.org/abs/2206.04658,MIT,NVIDIA,2022,1113wavefm,WaveFM (1-step),Distilled Flow Matching,24kHz Primary Benchmark,24000,19.53,70.7,3.672,0.963,48.51,7.26,3.71,3.74,3.85,0.0709,14.1,3065.6,Yes,False,https://github.com/lucas-ma/WaveFM,https://arxiv.org/abs/2406.00287,MIT,Ma et al.,2024,1214periodwave_24k,PeriodWave (30-step),Diffusion / Full SDE,24kHz Primary Benchmark,24000,29.81,70.6,4.147,0.964,126.28,10.73,4.15,4.21,2.65,0.7693,1.3,4469.1,No,False,https://github.com/kaist-dsp/PeriodWave,https://arxiv.org/abs/2408.06945,MIT,KAIST,2024,1315rfwave_libritts_24k,RFWave,Rectified Flow Matching,24kHz Primary Benchmark,24000,18.27,67.6,3.814,0.94,145.66,8.8,3.98,4.01,3.15,0.3752,2.7,11142.9,No,False,https://github.com/RF-Wave/RFWave,https://arxiv.org/abs/2406.18567,MIT,ByteDance / SJTU,2024,1416hifigan_universal_v1,HiFi-GAN (Universal V1),Time-domain MRF-GAN,22.05kHz Legacy Track,22050,13.94,64.8,3.019,0.947,71.28,7.7,3.45,3.5,4.45,0.0287,34.8,418.7,Yes,False,https://github.com/jik876/hifi-gan,https://arxiv.org/abs/2010.05646,MIT,Kakao Enterprise,2020,1517 