Automatic Speech Recognition on AISHELL-1 (test)
0.57CERFireRedASR2-AED
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FireRedASR2-AEDParameters=1B+, Streaming setting=non-streaming2026.03 | 0.57 | — | — | — | — | — | — | — | — | |
| NIM4-ASRModel Size=2.3B, Inference Mode=Offline2026.04 | 0.57 | — | — | — | — | — | — | — | — | |
| OursModel Size=2.3B2026.04 | 0.59 | — | — | — | — | — | — | — | — | |
| NIM4-ASRModel Size=2.3B, Inference Mode=Stream2026.04 | 0.6 | — | — | — | — | — | — | — | — | |
| FireRedASR2-LLMParameters=8B+, Streaming setting=non-streaming2026.03 | 0.64 | — | — | — | — | — | — | — | — | |
| FireRedASR2S LLMModel Size=8B+, Inference Mode=Offline2026.04 | 0.64 | — | — | — | — | — | — | — | — | |
| FireRedASR-LLMModel Size=8B+ (↑)2026.04 | 0.73 | — | — | — | — | — | — | — | — | |
| Step-Audio2-miniModel Size=8B+ (↑)2026.04 | 0.81 | — | — | — | — | — | — | — | — | |
| Step-Audio2 MiniModel Size=8B+, Inference Mode=Offline2026.04 | 0.81 | — | — | — | — | — | — | — | — | |
| Qwen3-Omni-InstModel Size=30B-A3B (↑)2026.04 | 0.92 | — | — | — | — | — | — | — | — | |
| Qwen3-Omni InstructModel Size=30B-A3B, Inference Mode=Offline2026.04 | 0.92 | — | — | — | — | — | — | — | — | |
| Qwen2.5-Omni-7BModel Parameters=7B2026.03 | 1.12 | — | — | — | — | — | — | — | — | |
| Qwen3-ASRType=open-source checkpoint, Streaming setting=non-streaming2026.03 | 1.48 | — | — | — | — | — | — | — | — | |
| Qwen3-ASR-1.7BModel Size=2.0B (↓)2026.04 | 1.51 | — | — | — | — | — | — | — | — | |
| Qwen3-ASR 1.7BModel Size=2.0B, Inference Mode=Offline2026.04 | 1.51 | — | — | — | — | — | — | — | — | |
| Doubao-ASRType=commercial API, Streaming setting=non-streaming, ITN and punctuation setting=disabled, VAD configuration=default2026.03 | 1.52 | — | — | — | — | — | — | — | — | |
| Fun-ASRType=commercial API, Streaming setting=non-streaming, ITN and punctuation setting=disabled, VAD configuration=default2026.03 | 1.64 | — | — | — | — | — | — | — | — | |
| Fun-ASR-nanoModel Size=0.8B (↓)2026.04 | 1.81 | — | — | — | — | — | — | — | — | |
| Fun-ASR NanoModel Size=0.8B, Inference Mode=Offline2026.04 | 1.81 | — | — | — | — | — | — | — | — | |
| Baichuan-Omni-1.5Model Parameters=7B2026.03 | 2.37 | — | — | — | — | — | — | — | — | |
| GLM-ASR-nanoModel Size=1.5B (↓)2026.04 | 2.41 | — | — | — | — | — | — | — | — | |
| GLM-ASR NanoModel Size=1.5B, Inference Mode=Offline2026.04 | 2.41 | — | — | — | — | — | — | — | — | |
| GLM-4-VoiceModel Parameters=9B2026.03 | 2.46 | — | — | — | — | — | — | — | — | |
| O_cpLanguage=Man2023.09 | 3.1 | — | — | — | — | — | — | — | — | |
| SE-WSBOLanguage Model usage=With LM2022.07 | 4.05 | — | — | — | — | — | — | — | — | |
| UOT-BERT-CTClambda1=0.5, lambda2=1.0, epsilon=0.05, eta=0.3, Decoding strategy=CTC greedy search-based2025.09 | 4.06 | — | — | — | — | — | — | — | — | |
| Neural TransducerLM=true, # Param=90 M2023.01 | 4.1 | — | — | — | — | — | — | — | — | |
| CIF + HKDLM=true, # Param=47 M2023.01 | 4.1 | — | — | — | — | — | — | — | — | |
| O_nbLanguage=Man2023.09 | 4.1 | — | — | — | — | — | — | — | — | |
| UOT-BERT-CTClambda1=1.0, lambda2=1.0, epsilon=0.05, eta=0.3, Decoding strategy=CTC greedy search-based2025.09 | 4.13 | — | — | — | — | — | — | — | — | |
| UOT-BERT-CTClambda1=1.0, lambda2=0.5, epsilon=0.05, eta=0.3, Decoding strategy=CTC greedy search-based2025.09 | 4.13 | — | — | — | — | — | — | — | — | |
| Neural Transducer+LFMMILanguage Model usage=With LM2022.07 | 4.18 | — | — | — | — | — | — | — | — | |
| OT-BERT-CTCDecoding strategy=CTC greedy search-based, eta=0.32025.09 | 4.19 | — | — | — | — | — | — | — | — | |
| CIF + HKDLM=false, # Param=47 M2023.01 | 4.2 | — | — | — | — | — | — | — | — | |
| UOT-BERT-CTClambda1=10.0, lambda2=10.0, epsilon=0.05, eta=0.3, Decoding strategy=CTC greedy search-based2025.09 | 4.21 | — | — | — | — | — | — | — | — | |
| Base-WSBOLanguage Model usage=With LM2022.07 | 4.22 | — | — | — | — | — | — | — | — | |
| K2Language Model usage=With LM2022.07 | 4.26 | — | — | — | — | — | — | — | — | |
| Base-WSBO+SoftmaxLanguage Model usage=With LM2022.07 | 4.28 | — | — | — | — | — | — | — | — | |
| Zipformer-LParams (M)=157.3, Type=pruned transducer2023.10 | 4.28 | — | — | — | — | — | — | — | — | |
| SE-WSBOLanguage Model usage=No LM2022.07 | 4.29 | — | — | — | — | — | — | — | — | |
| Icefall2023.01 | 4.3 | — | — | — | — | — | — | — | — | |
| WeNetLanguage Model usage=With LM2022.07 | 4.36 | — | — | — | — | — | — | — | — | |
| BranchformerLM=false, # Param=45 M2023.01 | 4.4 | — | — | — | — | — | — | — | — | |
| WeNet# Param=46 M2023.01 | 4.4 | — | — | — | — | — | — | — | — | |
| CIF + LRDLM=true, # Param=47 M2023.01 | 4.4 | — | — | — | — | — | — | — | — | |
| Zipformer-MParams (M)=73.4, Type=pruned transducer2023.10 | 4.4 | — | — | — | — | — | — | — | — | |
| BranchformerParams (M)=45.4, Type=CTC/AED2023.10 | 4.43 | — | — | — | — | — | — | — | — | |
| Branchformer (B), hybridType=AR, Params=45M, LM usage=without LM, Decoding beam size=102025.09 | 4.43 | — | — | — | — | — | — | — | — | |
| UMA-Split (prop.)Type=NAR, Params=46M, LM usage=without LM2025.09 | 4.43 | — | — | — | — | — | — | — | — | |
| Base-WSBOLanguage Model usage=No LM2022.07 | 4.48 | — | — | — | — | — | — | — | — | |
| CIF + LRDLM=false, # Param=47 M2023.01 | 4.5 | — | — | — | — | — | — | — | — | |
| E-Branchformer in ESPnetParams (M)=37.9, Type=CTC/AED2023.10 | 4.5 | — | — | — | — | — | — | — | — | |
| E-Branchformer, hybridType=AR, Params=57M, LM usage=without LM, Decoding beam size=102025.09 | 4.53 | — | — | — | — | — | — | — | — | |
| UMA-Split (w/o split module)Type=NAR, Params=45M, LM usage=without LM2025.09 | 4.53 | — | — | — | — | — | — | — | — | |
| Base-WSBO+SoftmaxLanguage Model usage=No LM2022.07 | 4.54 | — | — | — | — | — | — | — | — | |
| CIF + ACDLM=true, # Param=47 M2023.01 | 4.6 | — | — | — | — | — | — | — | — | |
| Hybrid (Conformer)Search=beam search, RTF=0.461, Params (M)=46.32023.09 | 4.6 | — | — | — | — | — | — | — | — | |
| WeNetLanguage Model usage=No LM2022.07 | 4.61 | — | — | — | — | — | — | — | — | |
| Conformer in WeNetParams (M)=46.3, Type=CTC/AED2023.10 | 4.61 | — | — | — | — | — | — | — | — | |
| EffectiveASR LargeType=NAR, Params=76M, LM usage=without LM2025.09 | 4.62 | — | — | — | — | — | — | — | — | |
| Zipformer-SParams (M)=30.2, Type=pruned transducer2023.10 | 4.67 | — | — | — | — | — | — | — | — | |
| LAStraining hours=18.12024.09 | 4.67 | — | — | — | — | — | — | — | — | |
| NAR-BERT-ASRDecoding strategy=CTC greedy search-based, eta=0.32025.09 | 4.68 | — | — | — | — | — | — | — | — | |
| EspnetLanguage Model usage=With LM2022.07 | 4.7 | — | — | — | — | — | — | — | — | |
| ESPnet ConformerLM=true, # Param=46 M2023.01 | 4.7 | — | — | — | — | — | — | — | — | |
| CIF + ACDLM=false, # Param=47 M2023.01 | 4.7 | — | — | — | — | — | — | — | — | |
| UMA + self-conditionRTF=0.045, Params (M)=44.72023.09 | 4.7 | — | — | — | — | — | — | — | — | |
| Original UMA ConformerType=NAR, Params=45M, LM usage=without LM2025.09 | 4.7 | — | — | — | — | — | — | — | — | |
| Lightweight Transducertraining hours=18.62024.09 | 4.76 | — | — | — | — | — | — | — | — | |
| CIFLM=true, # Param=47 M2023.01 | 4.8 | — | — | — | — | — | — | — | — | |
| UMARTF=0.039, Params (M)=42.62023.09 | 4.8 | — | — | — | — | — | — | — | — | |
| Zipformer-M, CTCType=NAR, Params=66M, LM usage=without LM2025.09 | 4.8 | — | — | — | — | — | — | — | — | |
| Conformer+CTC/AEDDecoding strategy=CTC greedy search-based, eta=0.32025.09 | 4.82 | — | — | — | — | — | — | — | — | |
| EspnetLanguage Model usage=No LM2022.07 | 4.9 | — | — | — | — | — | — | — | — | |
| ESPnet ConformerLM=false, # Param=46 M2023.01 | 4.9 | — | — | — | — | — | — | — | — | |
| CIFLM=false, # Param=47 M2023.01 | 4.9 | — | — | — | — | — | — | — | — | |
| Self-conditioned CTCRTF=0.059, Params (M)=51.52023.09 | 4.9 | — | — | — | — | — | — | — | — | |
| Conformer in ESPnetParams (M)=46.2, Type=CTC/AED2023.10 | 4.9 | — | — | — | — | — | — | — | — | |
| ProposedModel type=decoder-only, Streaming=false2026.01 | 4.9 | — | — | — | — | — | — | — | — | |
| Paraformer-v2 (S)Type=NAR, Params=50M, LM usage=without LM2025.09 | 4.9 | — | — | — | — | — | — | — | — | |
| CTC Infer w/o AED headType=NAR, Params=46M, LM usage=without LM2025.09 | 4.91 | — | — | — | — | — | — | — | — | |
| H2T-LoRALanguage=Man, Foundation Model=Chinese LlaMa2-7b, Training Pairs=20k Hypotheses-Transcription pairs2023.09 | 5 | — | — | — | — | — | — | — | — | |
| WeNet-U2Model type=encoder-decoder, Streaming=false2026.01 | 5 | — | — | — | — | — | — | — | — | |
| Transducertraining hours=64.22024.09 | 5.07 | — | — | — | — | — | — | — | — | |
| ProposedModel type=decoder-only, Streaming=true2026.01 | 5.1 | — | — | — | — | — | — | — | — | |
| Paraformer2023.09 | 5.2 | — | — | — | — | — | — | — | — | |
| Clean (No Attack)Defense=TI, Defense Type=Adversarial Training2026.06 | 5.22 | — | — | — | — | — | — | — | — | |
| BESTOWModel type=decoder-only, Streaming=true2026.01 | 5.3 | — | — | — | — | — | — | — | — | |
| U2params(M)=47, latency(ms)=320 + Δ2020.12 | 5.33 | — | — | — | — | — | — | — | — | |
| Muting WhisperDefense=TI, Defense Type=Adversarial Training2026.06 | 5.41 | — | — | — | — | — | — | — | — | |
| SPEECH-OMNI-LITEModel Parameters=32B, Projector-only training=true2026.03 | 5.46 | — | — | — | — | — | — | — | — | |
| Clean (No Attack)Defense=CE-AT, Defense Type=Adversarial Training2026.06 | 5.51 | — | — | — | — | — | — | — | — | |
| Hybrid (Conformer)RTF=0.125, Params (M)=46.32023.09 | 5.6 | — | — | — | — | — | — | — | — | |
| Clean (No Attack)Defense=DMW, Defense Type=Adversarial Training2026.06 | 5.66 | — | — | — | — | — | — | — | — | |
| Clean (No Attack)Defense=PVP Vote, Defense Type=Input Preprocessing2026.06 | 5.75 | — | — | — | — | — | — | — | — | |
| Conformer+CTCDecoding strategy=CTC greedy search-based, eta=0.32025.09 | 5.76 | — | — | — | — | — | — | — | — | |
| SPEECH-OMNI-LITEModel Parameters=8B2026.03 | 5.77 | — | — | — | — | — | — | — | — | |
| Clean (No Attack)Defense=DT-AT, Defense Type=Adversarial Training2026.06 | 5.87 | — | — | — | — | — | — | — | — | |
| Muting WhisperDefense=DMW, Defense Type=Adversarial Training2026.06 | 5.87 | — | — | — | — | — | — | — | — | |
| BTIModel type=decoder-only, Streaming=true2026.01 | 5.9 | — | — | — | — | — | — | — | — |