{"url":"/sota/speech-recognition-on-wsj-eval92","task":{"name":"Speech Recognition","url":"/task/speech-recognition","note":null},"dataset":{"name":"WSJ eval92","url":null},"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":17,"rows_with_code":8,"rows_with_paper_page":15,"rows_dated":15,"rows_using_additional_data":4},"rows":[{"rank_in_archive_order":1,"model":"Speechstew 100M","metrics":{"Word Error Rate (WER)":"1.3"},"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":2,"model":"ConformerXXL-P","metrics":{"Word Error Rate (WER)":"1.3"},"uses_additional_data":false,"paper_date":"2021-09-27","paper":"/paper/bigssl-exploring-the-frontier-of-large-scale","paper_url":"https://arxiv.org/abs/2109.13226v3","paper_title":"BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":3,"model":"Task activating prompting generative correction","metrics":{"Word Error Rate (WER)":"2.11"},"uses_additional_data":true,"paper_date":"2023-09-27","paper":"/paper/generative-speech-recognition-error","paper_url":"https://arxiv.org/abs/2309.15649v2","paper_title":"Generative Speech Recognition Error Correction with Large Language Models and Task-Activating Prompting","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":4,"model":"RobustGER","metrics":{"Word Error Rate (WER)":"2.2"},"uses_additional_data":true,"paper_date":"2024-02-08","paper":"/paper/it-s-never-too-late-fusing-acoustic","paper_url":"https://arxiv.org/abs/2402.05457v1","paper_title":"It's Never Too Late: Fusing Acoustic Information into Large Language Models for Automatic Speech Recognition","code":"https://github.com/Hypotheses-Paradise/UADF","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":4,"n_samples":6,"n_pointer_only_licence":0}},{"rank_in_archive_order":5,"model":"tdnn + chain","metrics":{"Word Error Rate (WER)":"2.32"},"uses_additional_data":false,"paper_date":"2016-09-08","paper":"/paper/purely-sequence-trained-neural-networks-for","paper_url":"https://www.danielpovey.com/files/2016_interspeech_mmi.pdf","paper_title":"Purely sequence-trained neural networks for ASR based on lattice-free MMI","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":6,"model":"CTC-CRF ST-NAS","metrics":{"Word Error Rate (WER)":"2.77"},"uses_additional_data":false,"paper_date":"2020-11-11","paper":"/paper/efficient-neural-architecture-search-for-end","paper_url":"https://arxiv.org/abs/2011.05649v1","paper_title":"Efficient Neural Architecture Search for End-to-end Speech Recognition via Straight-Through Gradients","code":"https://github.com/thu-spmi/ST-NAS","n_code_links":1,"syntology":null},{"rank_in_archive_order":7,"model":"End-to-end LF-MMI","metrics":{"Word Error Rate (WER)":"3.0"},"uses_additional_data":false,"paper_date":"2018-09-06","paper":"/paper/end-to-end-speech-recognition-using-lattice","paper_url":"https://www.isca-speech.org/archive/Interspeech_2018/abstracts/1423.html","paper_title":"End-to-end speech recognition using lattice-free MMI","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":8,"model":"Transformer with Relaxed Attention","metrics":{"Word Error Rate (WER)":"3.19"},"uses_additional_data":false,"paper_date":"2021-07-02","paper":"/paper/relaxed-attention-a-simple-method-to-boost","paper_url":"https://arxiv.org/abs/2107.01275v2","paper_title":"Relaxed Attention: A Simple Method to Boost Performance of End-to-End Automatic Speech Recognition","code":"https://github.com/freewym/espresso","n_code_links":1,"syntology":null},{"rank_in_archive_order":9,"model":"CTC-CRF VGG-BLSTM","metrics":{"Word Error Rate (WER)":"3.2"},"uses_additional_data":false,"paper_date":"2020-05-27","paper":"/paper/cat-a-ctc-crf-based-asr-toolkit-bridging-the","paper_url":"https://arxiv.org/abs/2005.13326v2","paper_title":"CAT: A CTC-CRF based ASR Toolkit Bridging the Hybrid and the End-to-end Approaches towards Data Efficiency and Low Latency","code":"https://github.com/thu-spmi/cat","n_code_links":1,"syntology":null},{"rank_in_archive_order":10,"model":"Espresso","metrics":{"Word Error Rate (WER)":"3.4"},"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":11,"model":"TC-DNN-BLSTM-DNN","metrics":{"Word Error Rate (WER)":"3.5"},"uses_additional_data":false,"paper_date":"2015-04-07","paper":"/paper/deep-recurrent-neural-networks-for-acoustic","paper_url":"http://arxiv.org/abs/1504.01482v1","paper_title":"Deep Recurrent Neural Networks for Acoustic Modelling","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":12,"model":"Convolutional Speech Recognition","metrics":{"Word Error Rate (WER)":"3.5"},"uses_additional_data":false,"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":13,"model":"test-set on open vocabulary (i.e. harder), model = HMM-DNN + pNorm*","metrics":{"Word Error Rate (WER)":"3.6"},"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":14,"model":"Deep Speech 2","metrics":{"Word Error Rate (WER)":"3.60"},"uses_additional_data":true,"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":15,"model":"CTC-CRF 4gram-LM","metrics":{"Word Error Rate (WER)":"3.79"},"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":16,"model":"CNN over RAW speech (wav)","metrics":{"Word Error Rate (WER)":"5.6"},"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":17,"model":"Jasper 10x3","metrics":{"Word Error Rate (WER)":"6.9"},"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}],"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 6,795 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":6795,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":2785},"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":3,"rows_with_any_sample_ran":3,"distinct_papers_with_graph_line":3,"distinct_papers_with_any_sample_ran":3,"samples_over_distinct_papers":{"n_ran":5,"n_unverified":41,"n_samples":46,"n_pointer_only_licence":1,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":5,"n_unverified":41,"n_samples":46,"n_pointer_only_licence":1,"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"}}}