{"url":"/sota/speech-recognition-on-librispeech-test-clean","task":{"name":"Speech Recognition","url":"/task/speech-recognition","note":null},"dataset":{"name":"LibriSpeech test-clean","url":"/dataset/librispeech"},"category":"Audio","categories":["Audio","Speech"],"category_note":null,"description":"**Speech Recognition** is the task of converting spoken language into text. It involves recognizing the words spoken in an audio recording and transcribing them into a written format. The goal is to accurately transcribe the speech in real-time or from recorded audio, taking into account factors such as accents, speaking speed, and background noise.\r\n\r\n<span style=\"color:grey; opacity: 0.6\">( Image credit: [SpecAugment](https://arxiv.org/pdf/1904.08779v2.pdf) )</span>","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["Word Error Rate (WER)"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Word Error Rate (WER)":"lower"}},"counts":{"rows":64,"rows_with_code":45,"rows_with_paper_page":60,"rows_dated":60,"rows_using_additional_data":24},"rows":[{"rank_in_archive_order":1,"model":"United Med ASR","metrics":{"Word Error Rate (WER)":"0.985"},"uses_additional_data":true,"paper_date":"2024-11-24","paper":"/paper/high-precision-medical-speech-recognition","paper_url":"https://arxiv.org/abs/2412.00055v1","paper_title":"High-precision medical speech recognition through synthetic data and semantic correction: UNITED-MEDASR","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":2,"model":"SAMBA ASR","metrics":{"Word Error Rate (WER)":"1.17"},"uses_additional_data":true,"paper_date":"2025-01-06","paper":"/paper/samba-asr-state-of-the-art-speech-recognition","paper_url":"https://arxiv.org/abs/2501.02832v3","paper_title":"Samba-ASR: State-Of-The-Art Speech Recognition Leveraging Structured State-Space Models","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":3,"model":"FAdam","metrics":{"Word Error Rate (WER)":"1.34"},"uses_additional_data":true,"paper_date":"2024-05-21","paper":"/paper/fadam-adam-is-a-natural-gradient-optimizer","paper_url":"https://arxiv.org/abs/2405.12807v10","paper_title":"FAdam: Adam is a natural gradient optimizer using diagonal empirical Fisher information","code":"https://github.com/lessw2020/fadam_pytorch","n_code_links":1,"syntology":null},{"rank_in_archive_order":4,"model":"Conformer + Wav2vec 2.0 + SpecAugment-based Noisy Student Training with Libri-Light","metrics":{"Word Error Rate (WER)":"1.4"},"uses_additional_data":true,"paper_date":"2020-10-20","paper":"/paper/pushing-the-limits-of-semi-supervised","paper_url":"https://arxiv.org/abs/2010.10504v2","paper_title":"Pushing the Limits of Semi-Supervised Learning for Automatic Speech Recognition","code":"https://github.com/tuanio/noisy-student-training-asr","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":2}},{"rank_in_archive_order":5,"model":"w2v-BERT XXL","metrics":{"Word Error Rate (WER)":"1.4"},"uses_additional_data":true,"paper_date":"2021-08-07","paper":"/paper/w2v-bert-combining-contrastive-learning-and","paper_url":"https://arxiv.org/abs/2108.06209v2","paper_title":"W2v-BERT: Combining Contrastive Learning and Masked Language Modeling for Self-Supervised Speech Pre-Training","code":"https://github.com/facebookresearch/fairseq/tree/ust/examples/w2vbert","n_code_links":4,"syntology":null},{"rank_in_archive_order":6,"model":"parakeet-rnnt-1.1b","metrics":{"Word Error Rate (WER)":"1.46"},"uses_additional_data":true,"paper_date":"2023-05-08","paper":"/paper/fast-conformer-with-linearly-scalable","paper_url":"https://arxiv.org/abs/2305.05084v6","paper_title":"Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":7,"model":"Conv + Transformer + wav2vec2.0 + pseudo labeling","metrics":{"Word Error Rate (WER)":"1.5"},"uses_additional_data":true,"paper_date":"2020-10-22","paper":"/paper/self-training-and-pre-training-are","paper_url":"https://arxiv.org/abs/2010.11430v1","paper_title":"Self-training and Pre-training are Complementary for Speech Recognition","code":"https://github.com/facebookresearch/fairseq","n_code_links":3,"syntology":null},{"rank_in_archive_order":8,"model":"ContextNet + SpecAugment-based Noisy Student Training with Libri-Light","metrics":{"Word Error Rate (WER)":"1.7"},"uses_additional_data":true,"paper_date":"2020-05-19","paper":"/paper/improved-noisy-student-training-for-automatic","paper_url":"https://arxiv.org/abs/2005.09629v2","paper_title":"Improved Noisy Student Training for Automatic Speech Recognition","code":"https://github.com/upskyy/ContextNet","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"SpeechStew (1B)","metrics":{"Word Error Rate (WER)":"1.7"},"uses_additional_data":true,"paper_date":"2021-04-05","paper":"/paper/speechstew-simply-mix-all-available-speech","paper_url":"https://arxiv.org/abs/2104.02133v3","paper_title":"SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":10,"model":"Multistream CNN with Self-Attentive SRU (WER includes text normalization)","metrics":{"Word Error Rate (WER)":"1.75"},"uses_additional_data":true,"paper_date":"2020-05-21","paper":"/paper/asapp-asr-multistream-cnn-and-self-attentive","paper_url":"https://arxiv.org/abs/2005.10469v1","paper_title":"ASAPP-ASR: Multistream CNN and Self-Attentive SRU for SOTA Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":11,"model":"Stateformer","metrics":{"Word Error Rate (WER)":"1.76"},"uses_additional_data":false,"paper_date":"2023-05-21","paper":"/paper/multi-head-state-space-model-for-speech","paper_url":"https://arxiv.org/abs/2305.12498v2","paper_title":"Multi-Head State Space Model for Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":12,"model":"wav2vec 2.0 with Libri-Light","metrics":{"Word Error Rate (WER)":"1.8"},"uses_additional_data":true,"paper_date":"2020-06-20","paper":"/paper/wav2vec-2-0-a-framework-for-self-supervised","paper_url":"https://arxiv.org/abs/2006.11477v3","paper_title":"wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations","code":"https://github.com/huggingface/transformers","n_code_links":25,"syntology":{"n_ran":2,"n_unverified":7,"n_samples":9,"n_pointer_only_licence":2}},{"rank_in_archive_order":13,"model":"HuBERT with Libri-Light","metrics":{"Word Error Rate (WER)":"1.8"},"uses_additional_data":true,"paper_date":"2021-06-14","paper":"/paper/hubert-self-supervised-speech-representation","paper_url":"https://arxiv.org/abs/2106.07447v1","paper_title":"HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units","code":"https://github.com/huggingface/transformers","n_code_links":11,"syntology":{"n_ran":0,"n_unverified":9,"n_samples":9,"n_pointer_only_licence":0}},{"rank_in_archive_order":14,"model":"WavLM Large","metrics":{"Word Error Rate (WER)":"1.8"},"uses_additional_data":false,"paper_date":"2021-10-26","paper":"/paper/wavlm-large-scale-self-supervised-pre","paper_url":"https://arxiv.org/abs/2110.13900v5","paper_title":"WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing","code":"https://github.com/microsoft/unilm","n_code_links":9,"syntology":null},{"rank_in_archive_order":15,"model":"E-Branchformer (L) + Internal Language Model Estimation","metrics":{"Word Error Rate (WER)":"1.81"},"uses_additional_data":false,"paper_date":"2022-09-30","paper":"/paper/e-branchformer-branchformer-with-enhanced","paper_url":"https://arxiv.org/abs/2210.00077v2","paper_title":"E-Branchformer: Branchformer with Enhanced merging for speech recognition","code":"https://github.com/espnet/espnet","n_code_links":1,"syntology":null},{"rank_in_archive_order":16,"model":"Zipformer+pruned transducer w/ CR-CTC (no  external language model)","metrics":{"Word Error Rate (WER)":"1.88"},"uses_additional_data":false,"paper_date":"2024-10-07","paper":"/paper/cr-ctc-consistency-regularization-on-ctc-for","paper_url":"https://arxiv.org/abs/2410.05101v4","paper_title":"CR-CTC: Consistency regularization on CTC for improved speech recognition","code":"https://github.com/k2-fsa/icefall","n_code_links":1,"syntology":null},{"rank_in_archive_order":17,"model":"ContextNet(L)","metrics":{"Word Error Rate (WER)":"1.9"},"uses_additional_data":false,"paper_date":"2020-05-07","paper":"/paper/contextnet-improving-convolutional-neural","paper_url":"https://arxiv.org/abs/2005.03191v3","paper_title":"ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context","code":"https://github.com/TensorSpeech/TensorFlowASR","n_code_links":6,"syntology":null},{"rank_in_archive_order":18,"model":"Conformer(L)","metrics":{"Word Error Rate (WER)":"1.9"},"uses_additional_data":false,"paper_date":"2020-05-16","paper":"/paper/conformer-convolution-augmented-transformer","paper_url":"https://arxiv.org/abs/2005.08100v1","paper_title":"Conformer: Convolution-augmented Transformer for Speech Recognition","code":"https://github.com/PaddlePaddle/PaddleSpeech","n_code_links":25,"syntology":{"n_ran":4,"n_unverified":3,"n_samples":7,"n_pointer_only_licence":2}},{"rank_in_archive_order":19,"model":"Transformer+Time reduction+Self Knowledge distillation","metrics":{"Word Error Rate (WER)":"1.9"},"uses_additional_data":false,"paper_date":"2021-03-17","paper":"/paper/transformer-based-asr-incorporating-time","paper_url":"https://arxiv.org/abs/2103.09903v1","paper_title":"Transformer-based ASR Incorporating Time-reduction Layer and Fine-tuning with Self-Knowledge Distillation","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":20,"model":"ContextNet(M)","metrics":{"Word Error Rate (WER)":"2"},"uses_additional_data":true,"paper_date":"2020-05-07","paper":"/paper/contextnet-improving-convolutional-neural","paper_url":"https://arxiv.org/abs/2005.03191v3","paper_title":"ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context","code":"https://github.com/TensorSpeech/TensorFlowASR","n_code_links":6,"syntology":null},{"rank_in_archive_order":21,"model":"Transformer Transducer","metrics":{"Word Error Rate (WER)":"2.0"},"uses_additional_data":false,"paper_date":"2020-11-05","paper":"/paper/improving-rnn-transducer-based-asr-with","paper_url":"https://arxiv.org/abs/2011.03109v2","paper_title":"Improving RNN Transducer Based ASR with Auxiliary Tasks","code":"https://github.com/upskyy/Transformer-Transducer","n_code_links":1,"syntology":null},{"rank_in_archive_order":22,"model":"Conformer(M)","metrics":{"Word Error Rate (WER)":"2"},"uses_additional_data":true,"paper_date":"2020-05-16","paper":"/paper/conformer-convolution-augmented-transformer","paper_url":"https://arxiv.org/abs/2005.08100v1","paper_title":"Conformer: Convolution-augmented Transformer for Speech Recognition","code":"https://github.com/PaddlePaddle/PaddleSpeech","n_code_links":25,"syntology":{"n_ran":4,"n_unverified":3,"n_samples":7,"n_pointer_only_licence":2}},{"rank_in_archive_order":23,"model":"SpeechStew (100M)","metrics":{"Word Error Rate (WER)":"2.0"},"uses_additional_data":false,"paper_date":"2021-04-05","paper":"/paper/speechstew-simply-mix-all-available-speech","paper_url":"https://arxiv.org/abs/2104.02133v3","paper_title":"SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":24,"model":"Qwen-Audio","metrics":{"Word Error Rate (WER)":"2.0"},"uses_additional_data":false,"paper_date":"2023-11-14","paper":"/paper/qwen-audio-advancing-universal-audio","paper_url":"https://arxiv.org/abs/2311.07919v2","paper_title":"Qwen-Audio: Advancing Universal Audio Understanding via Unified Large-Scale Audio-Language Models","code":"https://github.com/alibaba-damo-academy/FunASR","n_code_links":2,"syntology":{"n_ran":5,"n_unverified":2,"n_samples":7,"n_pointer_only_licence":7}},{"rank_in_archive_order":25,"model":"Zipformer+pruned transducer (no  external language model)","metrics":{"Word Error Rate (WER)":"2.00"},"uses_additional_data":false,"paper_date":"2023-10-17","paper":"/paper/zipformer-a-faster-and-better-encoder-for","paper_url":"https://arxiv.org/abs/2310.11230v4","paper_title":"Zipformer: A faster and better encoder for automatic speech recognition","code":"https://github.com/k2-fsa/icefall","n_code_links":1,"syntology":null},{"rank_in_archive_order":26,"model":"Zipformer+CR-CTC (no external language model)","metrics":{"Word Error Rate (WER)":"2.02"},"uses_additional_data":false,"paper_date":"2024-10-07","paper":"/paper/cr-ctc-consistency-regularization-on-ctc-for","paper_url":"https://arxiv.org/abs/2410.05101v4","paper_title":"CR-CTC: Consistency regularization on CTC for improved speech recognition","code":"https://github.com/k2-fsa/icefall","n_code_links":1,"syntology":null},{"rank_in_archive_order":27,"model":"Conv + Transformer AM + Pseudo-Labeling (ConvLM with Transformer Rescoring)","metrics":{"Word Error Rate (WER)":"2.03"},"uses_additional_data":false,"paper_date":"2019-11-19","paper":"/paper/end-to-end-asr-from-supervised-to-semi","paper_url":"https://arxiv.org/abs/1911.08460v3","paper_title":"End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures","code":"https://github.com/facebookresearch/wav2letter/tree/master/recipes/models/sota/2019","n_code_links":1,"syntology":null},{"rank_in_archive_order":28,"model":"Conv + Transformer AM + Iterative Pseudo-Labeling (n-gram LM + Transformer Rescoring)","metrics":{"Word Error Rate (WER)":"2.10"},"uses_additional_data":false,"paper_date":"2020-05-19","paper":"/paper/iterative-pseudo-labeling-for-speech","paper_url":"https://arxiv.org/abs/2005.09267v2","paper_title":"Iterative Pseudo-Labeling for Speech Recognition","code":"https://github.com/facebookresearch/wav2letter/tree/master/recipes/ipl","n_code_links":1,"syntology":null},{"rank_in_archive_order":29,"model":"CTC + Transformer LM rescoring","metrics":{"Word Error Rate (WER)":"2.10"},"uses_additional_data":false,"paper_date":"2020-05-19","paper":"/paper/fast-simpler-and-more-accurate-hybrid-asr","paper_url":"https://arxiv.org/abs/2005.09150v2","paper_title":"Faster, Simpler and More Accurate Hybrid ASR Systems Using Wordpieces","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":30,"model":"Conformer(S)","metrics":{"Word Error Rate (WER)":"2.1"},"uses_additional_data":false,"paper_date":"2020-05-16","paper":"/paper/conformer-convolution-augmented-transformer","paper_url":"https://arxiv.org/abs/2005.08100v1","paper_title":"Conformer: Convolution-augmented Transformer for Speech Recognition","code":"https://github.com/PaddlePaddle/PaddleSpeech","n_code_links":25,"syntology":{"n_ran":4,"n_unverified":3,"n_samples":7,"n_pointer_only_licence":2}},{"rank_in_archive_order":31,"model":"Branchformer + GFSA","metrics":{"Word Error Rate (WER)":"2.11"},"uses_additional_data":false,"paper_date":"2023-12-07","paper":"/paper/graph-convolutions-enrich-the-self-attention","paper_url":"https://arxiv.org/abs/2312.04234v5","paper_title":"Graph Convolutions Enrich the Self-Attention in Transformers!","code":"https://github.com/jeongwhanchoi/gfsa","n_code_links":1,"syntology":{"n_ran":19,"n_unverified":10,"n_samples":29,"n_pointer_only_licence":0}},{"rank_in_archive_order":32,"model":"Multi-Stream Self-Attention With Dilated 1D Convolutions","metrics":{"Word Error Rate (WER)":"2.20"},"uses_additional_data":false,"paper_date":"2019-10-01","paper":"/paper/state-of-the-art-speech-recognition-using","paper_url":"https://arxiv.org/abs/1910.00716v1","paper_title":"State-of-the-Art Speech Recognition Using Multi-Stream Self-Attention With Dilated 1D Convolutions","code":"https://github.com/s-omranpour/Pytorch-Speech-Recognition","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":33,"model":"LSTM Transducer","metrics":{"Word Error Rate (WER)":"2.23"},"uses_additional_data":true,"paper_date":"2021-04-07","paper":"/paper/librispeech-transducer-model-with-internal","paper_url":"https://arxiv.org/abs/2104.03006v2","paper_title":"Librispeech Transducer Model with Internal Language Model Prior Correction","code":"https://github.com/rwth-i6/returnn","n_code_links":2,"syntology":null},{"rank_in_archive_order":34,"model":"Hybrid + Transformer LM rescoring","metrics":{"Word Error Rate (WER)":"2.26"},"uses_additional_data":false,"paper_date":"2019-10-22","paper":"/paper/transformer-based-acoustic-modeling-for","paper_url":"https://arxiv.org/abs/1910.09799v2","paper_title":"Transformer-based Acoustic Modeling for Hybrid Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":35,"model":"Hybrid model with Transformer rescoring","metrics":{"Word Error Rate (WER)":"2.3"},"uses_additional_data":false,"paper_date":"2019-05-08","paper":"/paper/rwth-asr-systems-for-librispeech-hybrid-vs","paper_url":"https://arxiv.org/abs/1905.03072v3","paper_title":"RWTH ASR Systems for LibriSpeech: Hybrid vs Attention -- w/o Data Augmentation","code":"https://github.com/rwth-i6/returnn","n_code_links":2,"syntology":null},{"rank_in_archive_order":36,"model":"ContextNet(S)","metrics":{"Word Error Rate (WER)":"2.3"},"uses_additional_data":true,"paper_date":"2020-05-07","paper":"/paper/contextnet-improving-convolutional-neural","paper_url":"https://arxiv.org/abs/2005.03191v3","paper_title":"ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context","code":"https://github.com/TensorSpeech/TensorFlowASR","n_code_links":6,"syntology":null},{"rank_in_archive_order":37,"model":"Conv + Transformer AM (ConvLM  with Transformer Rescoring) (LS only)","metrics":{"Word Error Rate (WER)":"2.31"},"uses_additional_data":false,"paper_date":"2019-11-19","paper":"/paper/end-to-end-asr-from-supervised-to-semi","paper_url":"https://arxiv.org/abs/1911.08460v3","paper_title":"End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures","code":"https://github.com/facebookresearch/wav2letter/tree/master/recipes/models/sota/2019","n_code_links":1,"syntology":null},{"rank_in_archive_order":38,"model":"Squeezeformer (L)","metrics":{"Word Error Rate (WER)":"2.47"},"uses_additional_data":false,"paper_date":"2022-06-02","paper":"/paper/squeezeformer-an-efficient-transformer-for","paper_url":"https://arxiv.org/abs/2206.00888v2","paper_title":"Squeezeformer: An Efficient Transformer for Automatic Speech Recognition","code":"https://github.com/NVIDIA/NeMo/tree/main/examples/asr/conf/squeezeformer","n_code_links":4,"syntology":{"n_ran":31,"n_unverified":18,"n_samples":49,"n_pointer_only_licence":0}},{"rank_in_archive_order":39,"model":"LAS + SpecAugment","metrics":{"Word Error Rate (WER)":"2.5"},"uses_additional_data":true,"paper_date":"2019-04-18","paper":"/paper/specaugment-a-simple-data-augmentation-method","paper_url":"https://arxiv.org/abs/1904.08779v3","paper_title":"SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition","code":"https://github.com/mozilla/DeepSpeech","n_code_links":30,"syntology":{"n_ran":1,"n_unverified":17,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":40,"model":"Transformer","metrics":{"Word Error Rate (WER)":"2.6"},"uses_additional_data":true,"paper_date":"2019-09-13","paper":"/paper/a-comparative-study-on-transformer-vs-rnn-in","paper_url":"https://arxiv.org/abs/1909.06317v2","paper_title":"A Comparative Study on Transformer vs RNN in Speech Applications","code":"https://github.com/espnet/espnet","n_code_links":2,"syntology":null},{"rank_in_archive_order":41,"model":"QuartzNet15x5","metrics":{"Word Error Rate (WER)":"2.69"},"uses_additional_data":false,"paper_date":"2019-10-22","paper":"/paper/quartznet-deep-automatic-speech-recognition","paper_url":"https://arxiv.org/abs/1910.10261v1","paper_title":"QuartzNet: Deep Automatic Speech Recognition with 1D Time-Channel Separable Convolutions","code":"https://github.com/NVIDIA/NeMo","n_code_links":15,"syntology":{"n_ran":3,"n_unverified":9,"n_samples":12,"n_pointer_only_licence":2}},{"rank_in_archive_order":42,"model":"LAS (no LM)","metrics":{"Word Error Rate (WER)":"2.7"},"uses_additional_data":true,"paper_date":"2019-04-18","paper":"/paper/specaugment-a-simple-data-augmentation-method","paper_url":"https://arxiv.org/abs/1904.08779v3","paper_title":"SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition","code":"https://github.com/mozilla/DeepSpeech","n_code_links":30,"syntology":{"n_ran":1,"n_unverified":17,"n_samples":18,"n_pointer_only_licence":0}},{"rank_in_archive_order":43,"model":"wav2vec_wav2letter","metrics":{"Word Error Rate (WER)":"2.7"},"uses_additional_data":false,"paper_date":"2020-10-22","paper":"/paper/self-training-and-pre-training-are","paper_url":"https://arxiv.org/abs/2010.11430v1","paper_title":"Self-training and Pre-training are Complementary for Speech Recognition","code":"https://github.com/facebookresearch/fairseq","n_code_links":3,"syntology":null},{"rank_in_archive_order":44,"model":"Espresso","metrics":{"Word Error Rate (WER)":"2.8"},"uses_additional_data":false,"paper_date":"2019-09-18","paper":"/paper/espresso-a-fast-end-to-end-neural-speech","paper_url":"https://arxiv.org/abs/1909.08723v3","paper_title":"Espresso: A Fast End-to-end Neural Speech Recognition Toolkit","code":"https://github.com/freewym/espresso","n_code_links":1,"syntology":{"n_ran":1,"n_unverified":0,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":45,"model":"Jasper DR 10x5 (+ Time/Freq Masks)","metrics":{"Word Error Rate (WER)":"2.84"},"uses_additional_data":false,"paper_date":"2019-04-05","paper":"/paper/jasper-an-end-to-end-convolutional-neural","paper_url":"https://arxiv.org/abs/1904.03288v3","paper_title":"Jasper: An End-to-End Convolutional Neural Acoustic Model","code":"https://github.com/osmr/imgclsmob","n_code_links":10,"syntology":null},{"rank_in_archive_order":46,"model":"Jasper DR 10x5","metrics":{"Word Error Rate (WER)":"2.95"},"uses_additional_data":false,"paper_date":"2019-04-05","paper":"/paper/jasper-an-end-to-end-convolutional-neural","paper_url":"https://arxiv.org/abs/1904.03288v3","paper_title":"Jasper: An End-to-End Convolutional Neural Acoustic Model","code":"https://github.com/osmr/imgclsmob","n_code_links":10,"syntology":null},{"rank_in_archive_order":47,"model":"tdnn + chain + rnnlm rescoring","metrics":{"Word Error Rate (WER)":"3.06"},"uses_additional_data":false,"paper_date":"2018-04-15","paper":"/paper/neural-network-language-modeling-with-letter","paper_url":"https://www.cs.jhu.edu/~hxu/neural-network-language.pdf","paper_title":"Neural Network Language Modeling with Letter-based Features and Importance Sampling","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":48,"model":"Convolutional Speech Recognition","metrics":{"Word Error Rate (WER)":"3.26"},"uses_additional_data":true,"paper_date":"2018-12-17","paper":"/paper/fully-convolutional-speech-recognition","paper_url":"http://arxiv.org/abs/1812.06864v2","paper_title":"Fully Convolutional Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":49,"model":"MT4SSL","metrics":{"Word Error Rate (WER)":"3.4"},"uses_additional_data":false,"paper_date":"2022-11-14","paper":"/paper/mt4ssl-boosting-self-supervised-speech","paper_url":"https://arxiv.org/abs/2211.07321v3","paper_title":"MT4SSL: Boosting Self-Supervised Speech Representation Learning by Integrating Multiple Targets","code":"https://github.com/ddlbojack/mt4ssl","n_code_links":1,"syntology":null},{"rank_in_archive_order":50,"model":"Model Unit Exploration","metrics":{"Word Error Rate (WER)":"3.60"},"uses_additional_data":false,"paper_date":"2019-02-05","paper":"/paper/model-unit-exploration-for-sequence-to","paper_url":"https://arxiv.org/abs/1902.01955v2","paper_title":"On the Choice of Modeling Unit for Sequence-to-Sequence Speech Recognition","code":"https://github.com/30stomercury/Automatic_Speech_Recognition","n_code_links":3,"syntology":null},{"rank_in_archive_order":51,"model":"Seq-to-seq attention","metrics":{"Word Error Rate (WER)":"3.82"},"uses_additional_data":true,"paper_date":"2018-05-08","paper":"/paper/improved-training-of-end-to-end-attention","paper_url":"http://arxiv.org/abs/1805.03294v1","paper_title":"Improved training of end-to-end attention models for speech recognition","code":"https://github.com/rwth-i6/returnn","n_code_links":14,"syntology":{"n_ran":0,"n_unverified":12,"n_samples":12,"n_pointer_only_licence":0}},{"rank_in_archive_order":52,"model":"CTC-CRF 4gram-LM","metrics":{"Word Error Rate (WER)":"4.09"},"uses_additional_data":false,"paper_date":"2019-04-16","paper":"/paper/crf-based-single-stage-acoustic-modeling-with","paper_url":"https://ieeexplore.ieee.org/document/8682256","paper_title":"CRF-based Single-stage Acoustic Modeling with CTC Topology","code":"https://github.com/thu-spmi/cat","n_code_links":1,"syntology":null},{"rank_in_archive_order":53,"model":"HMM-TDNN trained with MMI + data augmentation (speed) + iVectors + 3 regularizations","metrics":{"Word Error Rate (WER)":"4.3"},"uses_additional_data":false,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":54,"model":"Centaurus (30 M)","metrics":{"Word Error Rate (WER)":"4.4"},"uses_additional_data":false,"paper_date":"2025-01-22","paper":"/paper/let-ssms-be-convnets-state-space-modeling","paper_url":"https://arxiv.org/abs/2501.13230v1","paper_title":"Let SSMs be ConvNets: State-space Modeling with Optimal Tensor Contractions","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":55,"model":"HMM-TDNN + iVectors","metrics":{"Word Error Rate (WER)":"4.8"},"uses_additional_data":true,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":56,"model":"Gated ConvNets","metrics":{"Word Error Rate (WER)":"4.8"},"uses_additional_data":false,"paper_date":"2017-12-22","paper":"/paper/letter-based-speech-recognition-with-gated","paper_url":"http://arxiv.org/abs/1712.09444v2","paper_title":"Letter-Based Speech Recognition with Gated ConvNets","code":"https://github.com/eric-erki/wav2letter","n_code_links":2,"syntology":null},{"rank_in_archive_order":57,"model":"Deep Speech 2","metrics":{"Word Error Rate (WER)":"5.33"},"uses_additional_data":false,"paper_date":"2015-12-08","paper":"/paper/deep-speech-2-end-to-end-speech-recognition","paper_url":"http://arxiv.org/abs/1512.02595v1","paper_title":"Deep Speech 2: End-to-End Speech Recognition in English and Mandarin","code":"https://github.com/tensorflow/models/tree/master/research/deep_speech","n_code_links":35,"syntology":{"n_ran":2,"n_unverified":37,"n_samples":39,"n_pointer_only_licence":0}},{"rank_in_archive_order":58,"model":"CTC + policy learning","metrics":{"Word Error Rate (WER)":"5.42"},"uses_additional_data":false,"paper_date":"2017-12-19","paper":"/paper/improving-end-to-end-speech-recognition-with-1","paper_url":"http://arxiv.org/abs/1712.07101v1","paper_title":"Improving End-to-End Speech Recognition with Policy Learning","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":59,"model":"HMM-DNN + pNorm*","metrics":{"Word Error Rate (WER)":"5.5"},"uses_additional_data":true,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":60,"model":"Li-GRU","metrics":{"Word Error Rate (WER)":"6.2"},"uses_additional_data":false,"paper_date":"2018-11-19","paper":"/paper/the-pytorch-kaldi-speech-recognition-toolkit","paper_url":"http://arxiv.org/abs/1811.07453v2","paper_title":"The PyTorch-Kaldi Speech Recognition Toolkit","code":"https://github.com/mravanelli/pytorch-kaldi","n_code_links":11,"syntology":{"n_ran":5,"n_unverified":1,"n_samples":6,"n_pointer_only_licence":6}},{"rank_in_archive_order":61,"model":"Snips","metrics":{"Word Error Rate (WER)":"6.4"},"uses_additional_data":false,"paper_date":"2018-05-25","paper":"/paper/snips-voice-platform-an-embedded-spoken","paper_url":"http://arxiv.org/abs/1805.10190v3","paper_title":"Snips Voice Platform: an embedded Spoken Language Understanding system for private-by-design voice interfaces","code":"https://github.com/snipsco/snips-nlu","n_code_links":16,"syntology":{"n_ran":1,"n_unverified":14,"n_samples":15,"n_pointer_only_licence":0}},{"rank_in_archive_order":62,"model":"Local Prior Matching (Large Model)","metrics":{"Word Error Rate (WER)":"7.19"},"uses_additional_data":false,"paper_date":"2020-02-24","paper":"/paper/semi-supervised-speech-recognition-via-local","paper_url":"https://arxiv.org/abs/2002.10336v1","paper_title":"Semi-Supervised Speech Recognition via Local Prior Matching","code":"https://github.com/facebookresearch/wav2letter","n_code_links":1,"syntology":null},{"rank_in_archive_order":63,"model":"HMM-(SAT)GMM","metrics":{"Word Error Rate (WER)":"8.0"},"uses_additional_data":true,"paper_date":null,"paper":null,"paper_url":null,"paper_title":"","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":64,"model":"AmNet","metrics":{"Word Error Rate (WER)":"8.6"},"uses_additional_data":false,"paper_date":"2021-08-03","paper":"/paper/amortized-neural-networks-for-low-latency","paper_url":"https://arxiv.org/abs/2108.01553v1","paper_title":"Amortized Neural Networks for Low-Latency Speech Recognition","code":null,"n_code_links":0,"syntology":null}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":18,"rows_with_any_sample_ran":16,"distinct_papers_with_graph_line":15,"distinct_papers_with_any_sample_ran":13,"samples_over_distinct_papers":{"n_ran":77,"n_unverified":139,"n_samples":216,"n_pointer_only_licence":23,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":86,"n_unverified":162,"n_samples":248,"n_pointer_only_licence":27,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}