List your company
ResearchFormerly WizWand

Automatic Speech Recognition on LibriSpeech (LS) clean

1.67WERWhisfusion + rerank
0.87686.230911.58516.9391Aug 9, 2025Sep 7, 2025Oct 7, 2025Nov 6, 2025Dec 6, 2025Jan 5, 2026Feb 4, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Whisfusion + rerankDecoding Architecture=NAR, Params (B)=2.42025.081.67—143.276.07
Qwen3-ASR-1.7BDecoding Architecture=AR, Params (B)=2.02025.081.69—20.125.62
WhisfusionDecoding Architecture=NAR, Params (B)=1.52025.081.74—173.456.55
Whisper-large-v3Decoding Architecture=AR, Params (B)=1.62025.081.97—35.337.11
Canary-1b-v2Decoding Architecture=AR, Params (B)=1.02025.082.09—47.876.4
Whisper-turboDecoding Architecture=AR, Params (B)=0.82025.082.1—143.998.86
Drax (MBR, 8/16)Decoding Architecture=NAR, Params (B)=1.22025.082.4—33.168.34
Whisper-LLaDADecoding Architecture=NAR, Params (B)=8.72025.082.41—7.929.82
OWSM-CTC v3.1Decoding Architecture=NAR, Params (B)=1.02025.082.46—665.748.27
Seamless ASRBackbone=Seamless2026.012.70.9——
UniAudio 2.0evaluation_setting=Seen-task evaluation2026.022.7———
SpeechMapper (EuroLLM)Backbone=EuroLLM, Stage=stage 2, Objective=ASR CE2026.012.91.1——
SpeechMapper (Llama 3.1)Backbone=Llama 3.1, Stage=stage 2, Objective=ASR CE2026.012.91.2——
SpeechMapper (EuroLLM)Backbone=EuroLLM, Stage=stage 2, Objective=CE+MSE2026.013.11.2——
MiMo-Audio-7B-Instructevaluation_setting=Seen-task evaluation2026.023.5———
MMS-allDecoding Architecture=NAR, Params (B)=1.02025.083.74—681.2112.93
Qwen2.5-Omni-7Bevaluation_setting=Seen-task evaluation2026.023.9———
BEST-IWSLT25-IFBackbone=BEST-IWSLT25-IF2026.014.21.7——
SpeechMapper (Llama 3.1)Backbone=Llama 3.1, Stage=stage 2, Objective=CE+MSE2026.014.92.7——
SpeechMapper (Llama 3.1)Backbone=Llama 3.1, Stage=stage 12026.019.46.5——
SpeechMapper (EuroLLM)Backbone=EuroLLM, Stage=stage 12026.0121.517.5——