# Speech Recognition on Librispeech (100 hours) and AISHELL v1

Source: [Speech Recognition on Librispeech (100 hours) and AISHELL v1](<https://www.sota2.com/research/sota/speech-recognition-on-librispeech-100-hours-and-aishell-v1>)

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

Data updated: 2026-02-26T06:08:37.983Z

Evaluation of Speech Recognition performance across two distinct datasets: the 100-hour subset of LibriSpeech (English) and the AISHELL v1 corpus (Chinese).

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) | Librispeech (100 hours) and AISHELL v1 |
| Metric | WER (English) |
| Unit | PERCENT |
| Direction | Lower is better |
| Results loaded | 5 of 5 |
| 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 (English)](<https://www.sota2.com/research/sota/speech-recognition-on-librispeech-100-hours-and-aishell-v1.md?metric=WER%20(English)>): Lower is better; unit: PERCENT.

- [WER (Chinese)](<https://www.sota2.com/research/sota/speech-recognition-on-librispeech-100-hours-and-aishell-v1.md?metric=WER%20(Chinese)>): Lower is better; unit: PERCENT.

- [Average WER](<https://www.sota2.com/research/sota/speech-recognition-on-librispeech-100-hours-and-aishell-v1.md?metric=Average%20WER>): Lower is better; unit: PERCENT.

## Recorded results

| Method | WER (English) | Result date | Report paper |
| --- | --- | --- | --- |
| FERERO-FT | 5.4 | Not recorded | 68fa5d2f1e209131ce9d7e68 |
| LS-FT | 6.8 | Not recorded | 68fa5d2f1e209131ce9d7e68 |
| Komatsu et al. | 7.11 | Not recorded | 68fa5d2f1e209131ce9d7e68 |
| Init. (M2ASR) | 7.3 | Not recorded | 68fa5d2f1e209131ce9d7e68 |
| w/o CPC | 11.8 | Not recorded | 68fa5d2f1e209131ce9d7e68 |

### FERERO-FT

WER (English): 5.4. Result ID: 698bba967a967eddb770e9f5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

[Implementation](<https://github.com/lisha-chen/FERERO/>)

Recorded evaluation configuration: {"fine-tuning\_mode":"FERERO","losses":"CTC losses + CPC loss"}. This record does not assert independent reproduction.

### LS-FT

WER (English): 6.8. Result ID: 698bba967a967eddb770e9f4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

Recorded evaluation configuration: {"fine-tuning\_mode":"Linear Scalarization","losses":"CTC losses + CPC loss"}. This record does not assert independent reproduction.

### Komatsu et al.

WER (English): 7.11. Result ID: 698bba967a967eddb770e9f1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa6c301e209131cea32b97.md>)

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

### Init. (M2ASR)

WER (English): 7.3. Result ID: 698bba967a967eddb770e9f3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

Recorded evaluation configuration: {"model\_state":"Initial pre-trained"}. This record does not assert independent reproduction.

### w/o CPC

WER (English): 11.8. Result ID: 698bba967a967eddb770e9f2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa5d2f1e209131ce9d7e68.md>)

Recorded evaluation configuration: {"optimization\_strategy":"Sum of supervised CTC losses"}. This record does not assert independent reproduction.

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