List your company
ResearchFormerly WizWand

Automatic Speech Recognition on LibriSpeech 960h clean (dev)

1.9WERHuBERT-L
1.81082.41293.0153.6171May 27, 2023Nov 27, 2023May 30, 2024Dec 1, 2024Jun 4, 2025Dec 6, 2025Jun 9, 2026
Updated 4mo ago

Evaluation results

MethodLinks
HuBERT-LTrain=CTC, Decode=Greedy2026.061.9
HuBERT-L + Llama 8B (4bit, r=32)Train=Joint, Decode=S2T2026.062
Whisper-L + Llama 8B (4bit, r=32)Train=Joint, Decode=S2T2026.062.3
Whisper-LDecode=Beam2026.062.4
HuBERT-L + Llama 8B (4bit, r=32)Train=Joint, Decode=PF-S2T2026.062.6
PDS-Base-32Architecture=Conformer, Group=E, L=12, dh=512, dff=2048, h=8, #Params=119M, Speedup=1.25x, Beam size=5, Batch size=50k2023.052.7
PDS-Base-8Architecture=Conformer, Group=E, L=12, dh=512, dff=2048, h=8, #Params=113M, Speedup=0.97x, Beam size=5, Batch size=50k2023.052.72
PDS-Base-16Architecture=Conformer, Group=E, L=12, dh=512, dff=2048, h=8, #Params=114M, Speedup=1.23x, Beam size=5, Batch size=50k2023.052.73
Stack-4 (Baseline)Architecture=Conformer, Group=E, L=12, dh=512, dff=2048, h=8, #Params=113M, Speedup=1.00x, Beam size=5, Batch size=50k2023.052.77
Whisper-L + Llama 8B (4bit, r=32)Train=Joint, Decode=PF-S2T2026.062.8
HuBERT-L + Llama 8B (4bit, r=32)Train=S2T, Decode=S2T2026.062.8
PDS-Base-8Architecture=Conformer, Group=D, L=12, dh=256, dff=2048, h=4, #Params=46M, Speedup=0.97x, Beam size=5, Batch size=100k2023.052.87
Stack-4 (wenet)Architecture=Conformer, Group=E, L=12, dh=512, dff=2048, h=8, #Params=109M, Implementation source=wenet [5], Beam size=5, Batch size=50k2023.052.9
AIPAArchitecture=Enc-CTC, Data Augmentation=SpecAugment + AIPA, Auxiliary Loss=InterCTC, Technique=COS2024.062.91
PDS-Base-16Architecture=Conformer, Group=D, L=12, dh=256, dff=2048, h=4, #Params=46M, Speedup=1.14x, Beam size=5, Batch size=100k2023.052.93
PDS-Base-32Architecture=Conformer, Group=D, L=12, dh=256, dff=2048, h=4, #Params=47M, Speedup=1.20x, Beam size=5, Batch size=100k2023.052.95
HuBERT-L + Llama 8B (4bit, r=32)Train=PF-S2T, Decode=PF-S2T2026.063
Stack-4 (Baseline)Architecture=Conformer, Group=D, L=12, dh=256, dff=2048, h=4, #Params=45M, Speedup=1.00x, Beam size=5, Batch size=100k2023.053.02
Whisper-L + Llama 8B (4bit, r=32)Train=S2T, Decode=S2T2026.063.1
Whisper-L + Llama 8B (4bit, r=32)Train=PF-S2T, Decode=PF-S2T2026.063.1
PDS-Deep-16Architecture=Transformer, Group=C, L=30, dh=256, dff=2048, h=4, #Params=54M, Speedup=1.19x, Beam size=5, Batch size=100k2023.053.15
PDS-Base-8Architecture=Transformer, Group=B, L=12, dh=512, dff=2048, h=8, #Params=75M, Speedup=1.08x, Beam size=5, Batch size=100k2023.053.17
Stack-4 (fairseq)Architecture=Transformer, Group=B, L=12, dh=512, dff=2048, h=8, #Params=71M, Implementation source=fairseq [4], Beam size=5, Batch size=100k2023.053.2
PDS-Deep-32Architecture=Transformer, Group=C, L=30, dh=256, dff=2048, h=4, #Params=55M, Speedup=1.27x, Beam size=5, Batch size=100k2023.053.26
PDS-Base-32Architecture=Transformer, Group=B, L=12, dh=512, dff=2048, h=8, #Params=82M, Speedup=1.47x, Beam size=5, Batch size=100k2023.053.32
PDS-Base-16Architecture=Transformer, Group=B, L=12, dh=512, dff=2048, h=8, #Params=76M, Speedup=1.34x, Beam size=5, Batch size=100k2023.053.34
PDS-Deep-8Architecture=Transformer, Group=C, L=30, dh=256, dff=2048, h=4, #Params=53M, Speedup=1.03x, Beam size=5, Batch size=100k2023.053.34
BaselineArchitecture=Enc-CTC, Data Augmentation=SpecAugment, Auxiliary Loss=InterCTC2024.063.47
Stack-4 (Baseline)Architecture=Transformer, Group=B, L=12, dh=512, dff=2048, h=8, #Params=71M, Speedup=1.00x, Beam size=5, Batch size=100k2023.053.53
PDS-Base-8Architecture=Transformer, Group=A, L=12, dh=256, dff=2048, h=4, #Params=30M, Speedup=0.99x, Beam size=5, Batch size=100k2023.053.57
PDS-Base-16Architecture=Transformer, Group=A, L=12, dh=256, dff=2048, h=4, #Params=30M, Speedup=1.14x, Beam size=5, Batch size=100k2023.053.71
Stack-4 (fairseq)Architecture=Transformer, Group=A, L=12, dh=256, dff=2048, h=4, #Params=30M, Implementation source=fairseq [4], Beam size=5, Batch size=100k2023.053.8
Stack-4 (Baseline)Architecture=Transformer, Group=C, L=30, dh=256, dff=2048, h=4, #Params=53M, Speedup=1.00x, Beam size=5, Batch size=100k2023.053.8
Stack-4 (Baseline)Architecture=Transformer, Group=A, L=12, dh=256, dff=2048, h=4, #Params=30M, Speedup=1.00x, Beam size=5, Batch size=100k2023.053.88
PDS-Base-32Architecture=Transformer, Group=A, L=12, dh=256, dff=2048, h=4, #Params=31M, Speedup=1.20x, Beam size=5, Batch size=100k2023.054.13