paper-with-me

Speech Recognition 벤치마크

Speech Recognition on AISHELL-1

20개 결과 · ⬇ CSV · JSON

Word Error Rate (WER) 낮을수록 좋음

0.55 5.087 9.625 14.16 18.7 2018-08 2026-11 Att — 18.7 (2018-08-30) CTC/Att — 6.7 (2019-09-13) CTC-CRF 4gram-LM — 6.34 (2020-05-27) U2 — 4.72 (2020-12-10) SE-WSBO With LM — 4.1 (2022-07-24) MMSpeech With LM — 1.9 (2022-11-29) CIF-HKD With LM — 4.1 (2023-01-30) BRA-E — 6.63 (2023-03-23) Paraformer-large — 1.95 (2023-05-18) Paraformer — 4.95 (2023-05-18) BAT — 4.97 (2023-05-19) UMA — 4.7 (2023-09-15) Qwen-Audio — 1.29 (2023-11-14) Seed-ASR — 0.68 (2024-07-05) Lightweight Transducer With LM — 4.03 (2024-09-05) Lightweight Transducer — 4.31 (2024-09-05) Zipformer+CR-CTC (no external language model) — 4.02 (2024-10-07) FireRedASR-AED — 0.55 (2025-01-24) Retrieval-Augmented Self-Taught Reasonin — 17.96 (2026-01-21) Latent — 8.4 (2026-08-02) Att — 18.7 (2018-08-30) CTC/Att — 6.7 (2019-09-13) CTC-CRF 4gram-LM — 6.34 (2020-05-27) U2 — 4.72 (2020-12-10) SE-WSBO With LM — 4.1 (2022-07-24) MMSpeech With LM — 1.9 (2022-11-29) Qwen-Audio — 1.29 (2023-11-14) Seed-ASR — 0.68 (2024-07-05) FireRedASR-AED — 0.55 (2025-01-24)
RankModel Word Error Rate (WER)Params(M) Extra Training Data PaperCodeYear
1 FireRedASR-AED 0.551,100 ✓ FireRedASR: Open-Source Industrial-Grade Mandarin Speech Recognition Models from Encoder-Decoder to LLM Integration fireredteam/fireredasr 2025
2 Seed-ASR 0.68– ✓ Seed-ASR: Understanding Diverse Speech and Contexts with LLM-based Speech Recognition 2024
3 Qwen-Audio 1.29– ✓ Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models alibaba-damo-academy/FunASR · qwenlm/qwen-audio 2023
4 MMSpeech With LM 1.9– MMSpeech: Multi-modal Multi-task Encoder-Decoder Pre-training for Speech Recognition ofa-sys/ofa 2022
5 Paraformer-large 1.95220 ✓ FunASR: A Fundamental End-to-End Speech Recognition Toolkit alibaba-damo-academy/FunASR 2023
6 Zipformer+CR-CTC (no external language model) 4.0266.2 CR-CTC: Consistency regularization on CTC for improved speech recognition k2-fsa/icefall 2024
7 Lightweight Transducer With LM 4.0345.3 Lightweight Transducer Based on Frame-Level Criterion wangmengzhi/Lightweight-Transducer 2024
8 SE-WSBO With LM 4.146 Improving Mandarin Speech Recogntion with Block-augmented Transformer LeonWlw/asr_blockformer · mininglamp-technology/asr-blockformer 2022
8 CIF-HKD With LM 4.147 Knowledge Transfer from Pre-trained Language Models to Cif-based Speech Recognizers via Hierarchical Distillation MingLunHan/CIF-PyTorch · minglunhan/cif-hieradist 2023
10 Lightweight Transducer 4.3145.3 Lightweight Transducer Based on Frame-Level Criterion wangmengzhi/Lightweight-Transducer 2024
11 UMA 4.744.7 Unimodal Aggregation for CTC-based Speech Recognition Audio-WestlakeU/UMA-ASR 2023
12 U2 4.7247 Unified Streaming and Non-streaming Two-pass End-to-end Model for Speech Recognition PaddlePaddle/PaddleSpeech · TeaPoly/Conformer-Athena · xianchao-wu/wenet-deep-sparse-conformer · +2 2020
13 Paraformer 4.9546.3 FunASR: A Fundamental End-to-End Speech Recognition Toolkit alibaba-damo-academy/FunASR 2023
14 BAT 4.9790 BAT: Boundary aware transducer for memory-efficient and low-latency ASR alibaba-damo-academy/FunASR 2023
15 CTC-CRF 4gram-LM 6.34– CAT: A CTC-CRF based ASR Toolkit Bridging the Hybrid and the End-to-end Approaches towards Data Efficiency and Low Latency thu-spmi/cat 2020
16 BRA-E 6.638.5 Beyond Universal Transformer: block reusing with adaptor in Transformer for automatic speech recognition 2023
17 CTC/Att 6.7– A Comparative Study on Transformer vs RNN in Speech Applications espnet/espnet · MindSpore-scientific-2/code-11 2019
18 Latent 자동 추출 8.4– Latent Softmax for Data-Efficient Phoneme-Based Multilingual ASR Across Tonal and Non-Tonal Languages 2026
19 Retrieval-Augmented Self-Taught Reasonin 자동 추출 17.96– Retrieval-Augmented Self-Taught Reasoning Model with Adaptive Chain-of-Thought for ASR Named Entity Correction 2026
20 Att 18.7– End-to-end Speech Recognition with Adaptive Computation Steps 2018
1–20 / 20 페이지당 10 20 50 100