Codesota · Speech · Speech Recognition · TED-LIUMTasks/Speech/Speech Recognition
Speech Recognition · benchmark dataset · 2018 · EN

TED-LIUM v3.

TED-LIUM is an English ASR corpus derived from public TED talks, with the v3 release providing ~452 hours of audio aligned to verbatim transcripts. Long-form prepared speech with diverse speakers, accents, and topics.

Paper ↗Submit a result ↵
§ 01 · Leaderboard

Best published scores.

50 results indexed across 1 metric. Shaded row marks current SOTA; ties broken by submission date.

Primary
wer · lower is better
wer· primary
50 rows
#ModelOrgSubmittedPaper / codewer
01Qwen3-ASR-1.7BOpenAlibabaJan 2026Qwen3-ASR Technical Report · code2.28
02Cohere Transcribe (Mar 2026)OpenCohereMar 2026pwc-dump2.49
03VibeVoice-ASR-HF—Jan 2026VIBEVOICE-ASR Technical Report2.57
04Canary-Qwen-2.5BOpenNVIDIAMar 2025Training and Inference Efficiency of Encoder-Decoder Spe…2.71
05Phi-4 Multimodal InstructOpenMicrosoftMar 2025Phi-4-Mini Technical Report: Compact yet Powerful Multim…2.94
06Granite Speech 4.1 2BOpenIBMMay 2025Granite-speech: open-source speech-aware LLMs with stron…3.07
07SYMPHONY-ASR—Jan 2026pwc-dump3.39
08Granite Speech 3.3 8BOpenIBMMay 2025Granite-speech: open-source speech-aware LLMs with stron…3.40
09Audio Flamingo 3—Jul 2025Audio Flamingo 3: Advancing Audio Intelligence with Full… · code3.50
10Whisper Large v3 TurboOpenOpenAIDec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code3.57
11Granite Speech 3.3 2BOpenIBMMay 2025Granite-speech: open-source speech-aware LLMs with stron…3.57
12Lite-whisper-large-v3-turbo-acc—Feb 2025LiteASR: Efficient Automatic Speech Recognition with Low… · code3.60
13Parakeet-ctc-0.6b—May 2023Fast Conformer with Linearly Scalable Attention for Effi…3.77
14Voxtral-Mini-4B-Realtime-2602OpenMistral AIFeb 2026Voxtral Realtime3.79
15Parakeet-rnnt-0.6b—May 2023Fast Conformer with Linearly Scalable Attention for Effi…3.85
16Distil-large-v3—Nov 2023Distil-Whisper: Robust Knowledge Distillation via Large-… · code3.86
17Whisper Large v3OpenOpenAIDec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code3.86
18Whisper Large v2OpenOpenAIDec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code3.90
19Whisper Large—Dec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code3.91
20Lite-whisper-large-v3-fast—Feb 2025LiteASR: Efficient Automatic Speech Recognition with Low… · code3.92
21Parakeet-tdt_ctc-110m—Apr 2023Efficient Sequence Transduction by Jointly Predicting To… · code4.07
22Whisper-small.en—Dec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code4.07
23Whisper-medium.en—Dec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code4.12
24Stt_en_conformer_ctc_large—May 2020Conformer: Convolution-augmented Transformer for Speech … · code4.41
25Stt_en_fastconformer_transducer_large—May 2023Fast Conformer with Linearly Scalable Attention for Effi…4.46
26Distil-small.en—Nov 2023Distil-Whisper: Robust Knowledge Distillation via Large-… · code4.54
27Stt_en_fastconformer_ctc_large—May 2023Fast Conformer with Linearly Scalable Attention for Effi…4.76
28Distil-medium.en—Nov 2023Distil-Whisper: Robust Knowledge Distillation via Large-… · code4.84
29Whisper-base.en—Dec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code4.87
30Distil-large-v2—Nov 2023Distil-Whisper: Robust Knowledge Distillation via Large-… · code4.87
31Asr-conformer-loquacious—Feb 2025pwc-dump4.98
32Moonshine-base—Oct 2024Moonshine: Speech Recognition for Live Transcription and… · code5.65
33Moonshine-tiny—Oct 2024Moonshine: Speech Recognition for Live Transcription and… · code5.69
34Niagara-38m-batch.en—Feb 2026pwc-dump5.71
35Whisper-tiny.en—Dec 2022Robust Speech Recognition via Large-Scale Weak Supervisi… · code5.97
36Moonshine-streaming-tiny—Jan 2026pwc-dump6.10
37Stt_en_conformer_ctc_small—May 2020Conformer: Convolution-augmented Transformer for Speech … · code7.16
38Asr-wav2vec2-librispeech—Jun 2021SpeechBrain: A General-Purpose Speech Toolkit · code7.58
39Wav2vec2-large-960h-lv60-self—Jun 2020wav2vec 2.0: A Framework for Self-Supervised Learning of… · code14.88
40Wav2vec2-conformer-rel-pos-large-960h-ft—Oct 2020fairseq S2T: Fast Speech-to-Text Modeling with fairseq · code15.77
41Hubert-xlarge-ls960-ft—Jun 2021HuBERT: Self-Supervised Speech Representation Learning b… · code15.83
42Data2vec-audio-large-960h—Feb 2022data2vec: A General Framework for Self-supervised Learni… · code15.86
43Wav2vec2-conformer-rope-large-960h-ft—Oct 2020fairseq S2T: Fast Speech-to-Text Modeling with fairseq · code15.92
44Hubert-large-ls960-ft—Jun 2021HuBERT: Self-Supervised Speech Representation Learning b… · code15.92
45Wav2vec2-large-robust-ft-libri-960h—Apr 2021Robust wav2vec 2.0: Analyzing Domain Shift in Self-Super… · code16.46
46Mms-1b-all—May 2023Scaling Speech Technology to 1,000+ Languages · code17.48
47wav2vec 2.0 Large (960h)OpenMeta AIJun 2020wav2vec 2.0: A Framework for Self-Supervised Learning of… · code18.85
48Data2vec-audio-base-960h—Feb 2022data2vec: A General Framework for Self-supervised Learni… · code19.49
49Wav2vec2-base-960h—Jun 2020wav2vec 2.0: A Framework for Self-Supervised Learning of… · code21.05
50Mms-1b-fl102—May 2023Scaling Speech Technology to 1,000+ Languages · code32.35
Fig 2 · Rows sorted by score within each metric. Shaded row marks SOTA. Dates reflect model or paper release where available, otherwise the date Codesota accessed the source.
§ 03 · Progress

6 steps
of state of the art.

Each row below marks a model that broke the previous record on wer. Intermediate submissions are kept in the leaderboard above; only SOTA-setting entries are re-listed here.

Lower scores win. Each subsequent entry improved upon the previous best.

SOTA line · wer
  1. May 16, 2020Stt_en_conformer_ctc_large4.41
  2. Dec 6, 2022Whisper Large v3 TurboOpenAI3.57
  3. Mar 3, 2025Phi-4 Multimodal InstructMicrosoft2.94
  4. Mar 7, 2025Canary-Qwen-2.5BNVIDIA2.71
  5. Jan 26, 2026VibeVoice-ASR-HF2.57
  6. Jan 29, 2026Qwen3-ASR-1.7BAlibaba2.28
Fig 3 · SOTA-setting models only. 6 entries span May 2020 → Jan 2026.
§ 04 · Literature

21 papers
tied to this benchmark.

Every paper below corresponds to at least one row in the leaderboard above. Click through for the arXiv preprint and, when available, the reference implementation.

§ 06 · Contribute

Have a score that beats
this table?

Submit a checkpoint and a reproduction script. We will run it, publish the score, and — if it takes the top — annotate the step on the progress chart with your name.

Submit a result ↵Read submission guide
What a submission needs
  • 01A public checkpoint or API endpoint
  • 02A reproduction script with frozen commit + seed
  • 03Declared evaluation environment (Python, deps)
  • 04One row per metric declared by this dataset
  • 05A contact so we can follow up on discrepancies