Speech Recognition On Librispeech Test Clean

评估指标

Word Error Rate (WER)

评测结果

各个模型在此基准测试上的表现结果

		Paper Title	Repository
AmNet	8.6	Amortized Neural Networks for Low-Latency Speech Recognition	-
HMM-(SAT)GMM	8.0	-	-
Local Prior Matching (Large Model)	7.19	Semi-Supervised Speech Recognition via Local Prior Matching
Snips	6.4	Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces
Li-GRU	6.2	The PyTorch-Kaldi Speech Recognition Toolkit
HMM-DNN + pNorm*	5.5	-	-
CTC + policy learning	5.42	Improving End-to-End Speech Recognition with Policy Learning	-
Deep Speech 2	5.33	Deep Speech 2: End-to-End Speech Recognition in English and Mandarin
Gated ConvNets	4.8	Letter-Based Speech Recognition with Gated ConvNets
HMM-TDNN + iVectors	4.8	-	-
Centaurus (30 M)	4.4	Let SSMs be ConvNets: State-space Modeling with Optimal Tensor Contractions	-
HMM-TDNN trained with MMI + data augmentation (speed) + iVectors + 3 regularizations	4.3	-	-
CTC-CRF 4gram-LM	4.09	CRF-based Single-stage Acoustic Modeling with CTC Topology	-
Seq-to-seq attention	3.82	Improved training of end-to-end attention models for speech recognition
Model Unit Exploration	3.60	On the Choice of Modeling Unit for Sequence-to-Sequence Speech Recognition
MT4SSL	3.4	MT4SSL: Boosting Self-Supervised Speech Representation Learning by Integrating Multiple Targets
Convolutional Speech Recognition	3.26	Fully Convolutional Speech Recognition	-
tdnn + chain + rnnlm rescoring	3.06	Neural Network Language Modeling with Letter-based Features and Importance Sampling	-
Jasper DR 10x5	2.95	Jasper: An End-to-End Convolutional Neural Acoustic Model
Jasper DR 10x5 (+ Time/Freq Masks)	2.84	Jasper: An End-to-End Convolutional Neural Acoustic Model

0 of 64 row(s) selected.