الرئيسية الأخبار أحدث الأوراق البحثية الدروس مجموعات البيانات الموسوعة SOTA نماذج LLM لوحة الأداء GPU الفعاليات

العربية

Speech Recognition On Aishell 1

المقاييس

Params(M)

Word Error Rate (WER)

النتائج

نتائج أداء النماذج المختلفة على هذا المعيار القياسي

اسم النموذج	Params(M)	Word Error Rate (WER)	Paper Title	Repository
U2	47	4.72	Unified Streaming and Non-streaming Two-pass End-to-end Model for Speech Recognition
Zipformer+CR-CTC (no external language model)	66.2	4.02	CR-CTC: Consistency regularization on CTC for improved speech recognition
Paraformer	46.3	4.95	FunASR: A Fundamental End-to-End Speech Recognition Toolkit
Qwen-Audio	-	1.29	Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models
Lightweight Transducer With LM	45.3	4.03	Lightweight Transducer Based on Frame-Level Criterion
Paraformer-large	220	1.95	FunASR: A Fundamental End-to-End Speech Recognition Toolkit
Att	-	18.7	End-to-end Speech Recognition with Adaptive Computation Steps	-
SE-WSBO With LM	46	4.1	Improving Mandarin Speech Recogntion with Block-augmented Transformer
CTC/Att	-	6.7	A Comparative Study on Transformer vs RNN in Speech Applications
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
UMA	44.7	4.7	Unimodal Aggregation for CTC-based Speech Recognition
MMSpeech With LM	-	1.9	MMSpeech: Multi-modal Multi-task Encoder-Decoder Pre-training for Speech Recognition
CIF-HKD With LM	47	4.1	Knowledge Transfer from Pre-trained Language Models to Cif-based Speech Recognizers via Hierarchical Distillation
BRA-E	8.5	6.63	Beyond Universal Transformer: block reusing with adaptor in Transformer for automatic speech recognition	-
FireRedASR-AED	1,100	0.55	FireRedASR: Open-Source Industrial-Grade Mandarin Speech Recognition Models from Encoder-Decoder to LLM Integration	-
Seed-ASR	-	0.68	Seed-ASR: Understanding Diverse Speech and Contexts with LLM-based Speech Recognition	-
Lightweight Transducer	45.3	4.31	Lightweight Transducer Based on Frame-Level Criterion
BAT	90	4.97	BAT: Boundary aware transducer for memory-efficient and low-latency ASR

0 of 18 row(s) selected.