# Speech Recognition on WSJ nov92 (test)

Source: [Speech Recognition on WSJ nov92 (test)](<https://www.sota2.com/research/sota/speech-recognition-on-wsj-nov92-test>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-26T06:08:26.564Z

Evaluation of Speech Recognition performance on the test split of the Wall Street Journal 1992 (nov92) corpus, typically measured using Word Error Rate (WER) and Letter Error Rate (LER).

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Speech Recognition |
| Dataset (including any stated split) | WSJ nov92 (test) |
| Metric | WER |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 34 of 34 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [WER](<https://www.sota2.com/research/sota/speech-recognition-on-wsj-nov92-test.md?metric=WER>): Lower is better; unit: PERCENT.

- [LER](<https://www.sota2.com/research/sota/speech-recognition-on-wsj-nov92-test.md?metric=LER>): Lower is better; unit: PERCENT.

## Recorded results

| Method | WER | Result date | Report paper |
| --- | --- | --- | --- |
| Fast Conformer | 1.42 | Not recorded | 68fa70bd1e209131cea4cf97 |
| Conformer | 1.47 | Not recorded | 68fa70bd1e209131cea4cf97 |
| Fast Conformer | 1.59 | Not recorded | 68fa70bd1e209131cea4cf97 |
| Conformer | 1.7 | Not recorded | 68fa70bd1e209131cea4cf97 |
| vq-wav2vec Gumbel + BERT base | 2.34 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec large | 2.43 | Not recorded | 68fa66551e209131cea0f8cc |
| wav2vec | 2.43 | Not recorded | 68fa67041e209131cea13caa |
| Supervised transfer-learning | 2.53 | Not recorded | 68fa66551e209131cea0f8cc |
| wav2vec large | 2.53 | Not recorded | 68fa66551e209131cea0f8cc |
| Supervised transfer-learning | 2.53 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec | 2.78 | Not recorded | 68fa66551e209131cea0f8cc |
| Lattice-free MMI | 2.8 | Not recorded | 68fa66551e209131cea0f8cc |
| Lattice-free MMI | 2.8 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec | 2.87 | Not recorded | 68fa66551e209131cea0f8cc |
| Deep Speech 2 | 3.1 | Not recorded | 68fa66551e209131cea0f8cc |
| Deep Speech 2 | 3.1 | Not recorded | 68fa67041e209131cea13caa |
| Log-mel filterbanks (Baseline) | 3.46 | Not recorded | 68fa66551e209131cea0f8cc |
| Baseline (log-mel) | 3.46 | Not recorded | 68fa67041e209131cea13caa |
| Trainable frontend | 3.5 | Not recorded | 68fa66551e209131cea0f8cc |
| Trainable frontend | 3.5 | Not recorded | 68fa67041e209131cea13caa |
| Log-mel filterbanks (Baseline) | 3.6 | Not recorded | 68fa66551e209131cea0f8cc |
| vq-wav2vec Gumbel + BERT base | 3.62 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec large | 4.32 | Not recorded | 68fa66551e209131cea0f8cc |
| wav2vec | 4.32 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec | 4.57 | Not recorded | 68fa66551e209131cea0f8cc |
| wav2vec | 4.61 | Not recorded | 68fa66551e209131cea0f8cc |
| wav2vec | 5.55 | Not recorded | 68fa66551e209131cea0f8cc |
| Log-mel filterbanks (Baseline) | 5.64 | Not recorded | 68fa66551e209131cea0f8cc |
| Baseline (log-mel) | 5.64 | Not recorded | 68fa67041e209131cea13caa |
| vq-wav2vec Gumbel | 6.1 | Not recorded | 68fa67041e209131cea13caa |
| vq-wav2vec Gumbel + BERT base | 9.39 | Not recorded | 68fa67041e209131cea13caa |
| wav2vec | 11.2 | Not recorded | 68fa67041e209131cea13caa |
| Baseline (log-mel) | 13.93 | Not recorded | 68fa67041e209131cea13caa |
| vq-wav2vec Gumbel | 14.67 | Not recorded | 68fa67041e209131cea13caa |

### Fast Conformer

WER: 1.42. Result ID: 699156843a6c314849076d1f.

[Reporting paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Implementation](<https://github.com/NVIDIA/NeMo>)

Recorded evaluation configuration: {"Decoder":"RNNT"}. This record does not assert independent reproduction.

### Conformer

WER: 1.47. Result ID: 699156843a6c314849076d1e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa678b1e209131cea16d0c.md>)

Recorded evaluation configuration: {"Decoder":"RNNT"}. This record does not assert independent reproduction.

### Fast Conformer

WER: 1.59. Result ID: 699156843a6c314849076d21.

[Reporting paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Implementation](<https://github.com/NVIDIA/NeMo>)

Recorded evaluation configuration: {"Decoder":"CTC"}. This record does not assert independent reproduction.

### Conformer

WER: 1.7. Result ID: 699156843a6c314849076d20.

[Reporting paper](<https://www.sota2.com/research/paper/68fa70bd1e209131cea4cf97.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa678b1e209131cea16d0c.md>)

Recorded evaluation configuration: {"Decoder":"CTC"}. This record does not assert independent reproduction.

### vq-wav2vec Gumbel + BERT base

WER: 2.34. Result ID: 698d14788b517eb4331ed808.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Implementation](<http://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM"}. This record does not assert independent reproduction.

### wav2vec large

WER: 2.43. Result ID: 698ce9a18b517eb4331d9f3e.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### wav2vec

WER: 2.43. Result ID: 698d14788b517eb4331ed807.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM"}. This record does not assert independent reproduction.

### Supervised transfer-learning

WER: 2.53. Result ID: 698ce9a18b517eb4331d9f33.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

Recorded evaluation configuration: {"phoneme-based":true}. This record does not assert independent reproduction.

### wav2vec large

WER: 2.53. Result ID: 698ce9a18b517eb4331d9f3b.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"WORD CONVLM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### Supervised transfer-learning

WER: 2.53. Result ID: 698d14788b517eb4331ed7fd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### wav2vec

WER: 2.78. Result ID: 698ce9a18b517eb4331d9f3d.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### Lattice-free MMI

WER: 2.8. Result ID: 698ce9a18b517eb4331d9f32.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

Recorded evaluation configuration: {"phoneme-based":true}. This record does not assert independent reproduction.

### Lattice-free MMI

WER: 2.8. Result ID: 698d14788b517eb4331ed7fc.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### wav2vec

WER: 2.87. Result ID: 698ce9a18b517eb4331d9f3a.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"WORD CONVLM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### Deep Speech 2

WER: 3.1. Result ID: 698ce9a18b517eb4331d9f30.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62321e209131ce9f82cd.md>)

Recorded evaluation configuration: {"labeled speech":"12K h"}. This record does not assert independent reproduction.

### Deep Speech 2

WER: 3.1. Result ID: 698d14788b517eb4331ed7fa.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62321e209131ce9f82cd.md>)

Recorded evaluation configuration: {"Training":"12K h labeled speech"}. This record does not assert independent reproduction.

### Log-mel filterbanks (Baseline)

WER: 3.46. Result ID: 698ce9a18b517eb4331d9f3c.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65691e209131cea0a2cb.md>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM"}. This record does not assert independent reproduction.

### Baseline (log-mel)

WER: 3.46. Result ID: 698d14788b517eb4331ed806.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62bd1e209131ce9fb0c3.md>)

Recorded evaluation configuration: {"Language Model":"CHAR CONVLM"}. This record does not assert independent reproduction.

### Trainable frontend

WER: 3.5. Result ID: 698ce9a18b517eb4331d9f31.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa642c1e209131cea03226.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Trainable frontend

WER: 3.5. Result ID: 698d14788b517eb4331ed7fb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa642c1e209131cea03226.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

### Log-mel filterbanks (Baseline)

WER: 3.6. Result ID: 698ce9a18b517eb4331d9f39.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65691e209131cea0a2cb.md>)

Recorded evaluation configuration: {"Language Model":"WORD CONVLM"}. This record does not assert independent reproduction.

### vq-wav2vec Gumbel + BERT base

WER: 3.62. Result ID: 698d14788b517eb4331ed805.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Implementation](<http://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-gram LM"}. This record does not assert independent reproduction.

### wav2vec large

WER: 4.32. Result ID: 698ce9a18b517eb4331d9f38.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-GRAM LM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### wav2vec

WER: 4.32. Result ID: 698d14788b517eb4331ed803.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-gram LM"}. This record does not assert independent reproduction.

### wav2vec

WER: 4.57. Result ID: 698ce9a18b517eb4331d9f36.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-GRAM LM","Pre-training Data":"Librispeech","Pre-training Hours":"960h"}. This record does not assert independent reproduction.

### wav2vec

WER: 4.61. Result ID: 698ce9a18b517eb4331d9f37.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-GRAM LM","Pre-training Data":"Libri + WSJ","Pre-training Hours":"1,041 h"}. This record does not assert independent reproduction.

### wav2vec

WER: 5.55. Result ID: 698ce9a18b517eb4331d9f35.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-GRAM LM","Pre-training Data":"Librispeech","Pre-training Hours":"80h"}. This record does not assert independent reproduction.

### Log-mel filterbanks (Baseline)

WER: 5.64. Result ID: 698ce9a18b517eb4331d9f34.

[Reporting paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa65691e209131cea0a2cb.md>)

Recorded evaluation configuration: {"Language Model":"4-GRAM LM"}. This record does not assert independent reproduction.

### Baseline (log-mel)

WER: 5.64. Result ID: 698d14788b517eb4331ed802.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62bd1e209131ce9fb0c3.md>)

Recorded evaluation configuration: {"Language Model":"4-gram LM"}. This record does not assert independent reproduction.

### vq-wav2vec Gumbel

WER: 6.1. Result ID: 698d14788b517eb4331ed804.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Implementation](<http://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"4-gram LM"}. This record does not assert independent reproduction.

### vq-wav2vec Gumbel + BERT base

WER: 9.39. Result ID: 698d14788b517eb4331ed801.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Implementation](<http://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"None"}. This record does not assert independent reproduction.

### wav2vec

WER: 11.2. Result ID: 698d14788b517eb4331ed7ff.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa66551e209131cea0f8cc.md>)

[Implementation](<https://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"None"}. This record does not assert independent reproduction.

### Baseline (log-mel)

WER: 13.93. Result ID: 698d14788b517eb4331ed7fe.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa62bd1e209131ce9fb0c3.md>)

Recorded evaluation configuration: {"Language Model":"None"}. This record does not assert independent reproduction.

### vq-wav2vec Gumbel

WER: 14.67. Result ID: 698d14788b517eb4331ed800.

[Reporting paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67041e209131cea13caa.md>)

[Implementation](<http://github.com/pytorch/fairseq>)

Recorded evaluation configuration: {"Language Model":"None"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
