Speech Recognition On Librispeech Test Clean

評価指標

Word Error Rate (WER)

評価結果

このベンチマークにおける各モデルのパフォーマンス結果

モデル名	Word Error Rate (WER)	Paper Title	Repository
LAS (no LM)	2.7	SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
CTC + Transformer LM rescoring	2.10	Faster, Simpler and More Accurate Hybrid ASR Systems Using Wordpieces	-
Stateformer	1.76	Multi-Head State Space Model for Speech Recognition	-
Conformer(L)	1.9	Conformer: Convolution-augmented Transformer for Speech Recognition
Hybrid model with Transformer rescoring	2.3	RWTH ASR Systems for LibriSpeech: Hybrid vs Attention -- w/o Data Augmentation
Conv + Transformer AM + Iterative Pseudo-Labeling (n-gram LM + Transformer Rescoring)	2.10	Iterative Pseudo-Labeling for Speech Recognition	-
Zipformer+CR-CTC (no external language model)	2.02	CR-CTC: Consistency regularization on CTC for improved speech recognition
Seq-to-seq attention	3.82	Improved training of end-to-end attention models for speech recognition
tdnn + chain + rnnlm rescoring	3.06	Neural Network Language Modeling with Letter-based Features and Importance Sampling	-
Transformer+Time reduction+Self Knowledge distillation	1.9	Transformer-based ASR Incorporating Time-reduction Layer and Fine-tuning with Self-Knowledge Distillation	-
Li-GRU	6.2	The PyTorch-Kaldi Speech Recognition Toolkit
Transformer	2.6	A Comparative Study on Transformer vs RNN in Speech Applications
Snips	6.4	Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces
w2v-BERT XXL	1.4	W2v-BERT: Combining Contrastive Learning and Masked Language Modeling for Self-Supervised Speech Pre-Training
Zipformer+pruned transducer w/ CR-CTC (no external language model)	1.88	CR-CTC: Consistency regularization on CTC for improved speech recognition
Conv + Transformer AM (ConvLM with Transformer Rescoring) (LS only)	2.31	End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures
Multi-Stream Self-Attention With Dilated 1D Convolutions	2.20	State-of-the-Art Speech Recognition Using Multi-Stream Self-Attention With Dilated 1D Convolutions
Zipformer+pruned transducer (no external language model)	2.00	Zipformer: A faster and better encoder for automatic speech recognition
United-MedASR (764M)	0.985	High-precision medical speech recognition through synthetic data and semantic correction: UNITED-MEDASR	-
ContextNet(M)	2	ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context

0 of 64 row(s) selected.