{ "id": 335411, "title": "Non-autoregressive error correction method for automatic speech recognition", "url": "/competitions/dlsprint/discussion/335411", "author_name": "Mobassir", "post_date": "2022-07-06T02:53:00.270000", "votes": 7, "comment_count": 0, "views": 0, "content": "

NeuralSpeech is a research project in Microsoft Research Asia focusing on neural network based speech processing, including automatic speech recognition (ASR), text to speech (TTS), etc.

\n

Currently this repo covers several research work:

\n

Automatic Speech Recognition

\n
    \n
  1. FastCorrect, NeurIPS 2021
    2. \n
  2. FastCorrect 2, EMNLP 2021 Findings
    3. \n
  3. CMatch for ASR, INTERSPEECH 2021
    4. \n
  4. Adapter for ASR, TASLP 2022
    5. \n
\n

Text to Speech

\n
    \n
  1. LightSpeech, ICASSP 2021
    2. \n
  2. PriorGrad, ICLR 2022
    3. \n
\n

For more research in NeuralSpeech project, you can refer to this page: https://speechresearch.github.io/. they will release more research work in the future.

\n

FastCorrect and FastCorrect 2 are worth exploring, i guess. FastCorrect paper link : https://arxiv.org/abs/2105.03842

\n

FastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition, by Yichong Leng, Xu Tan, Linchen Zhu, Jin Xu, Renqian Luo, Linquan Liu, Tao Qin, Xiang-Yang Li, Ed Lin and Tie-Yan Liu in NeurIPS 2021, is a novel non-autoregressive error correction method for automatic speech recognition. It use an encoder and a length predictor to detect errors and a decoder to correct errors.

\n

\"\"

\n

Edit alignment of FastCorrect
\n\"\"

\n

The whole pipeline of FastCorrect consists of 4 steps, you can find the details here : https://github.com/microsoft/NeuralSpeech/tree/master/FastCorrect

\n

another interesting paper work is titled CROSS-MODAL ASR POST-PROCESSING SYSTEM FOR ERROR CORRECTION AND UTTERANCE REJECTION which can be found here : https://arxiv.org/pdf/2201.03313.pdf

\n

few more related interesting works can be found here : https://github.com/facebookresearch/fairseq/blob/main/examples/nonautoregressive_translation/README.md

\n

maybe this discussion about spell correction can help you too : https://discuss.huggingface.co/t/asr-spell-correction/5103?page=2

", "messages": [ { "id": 1845047, "postDate": "2022-07-06T02:53:00.270Z", "content": "

NeuralSpeech is a research project in Microsoft Research Asia focusing on neural network based speech processing, including automatic speech recognition (ASR), text to speech (TTS), etc.

\n

Currently this repo covers several research work:

\n

Automatic Speech Recognition

\n
    \n
  1. FastCorrect, NeurIPS 2021
    2. \n
  2. FastCorrect 2, EMNLP 2021 Findings
    3. \n
  3. CMatch for ASR, INTERSPEECH 2021
    4. \n
  4. Adapter for ASR, TASLP 2022
    5. \n
\n

Text to Speech

\n
    \n
  1. LightSpeech, ICASSP 2021
    2. \n
  2. PriorGrad, ICLR 2022
    3. \n
\n

For more research in NeuralSpeech project, you can refer to this page: https://speechresearch.github.io/. they will release more research work in the future.

\n

FastCorrect and FastCorrect 2 are worth exploring, i guess. FastCorrect paper link : https://arxiv.org/abs/2105.03842

\n

FastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition, by Yichong Leng, Xu Tan, Linchen Zhu, Jin Xu, Renqian Luo, Linquan Liu, Tao Qin, Xiang-Yang Li, Ed Lin and Tie-Yan Liu in NeurIPS 2021, is a novel non-autoregressive error correction method for automatic speech recognition. It use an encoder and a length predictor to detect errors and a decoder to correct errors.

\n

\"\"

\n

Edit alignment of FastCorrect
\n\"\"

\n

The whole pipeline of FastCorrect consists of 4 steps, you can find the details here : https://github.com/microsoft/NeuralSpeech/tree/master/FastCorrect

\n

another interesting paper work is titled CROSS-MODAL ASR POST-PROCESSING SYSTEM FOR ERROR CORRECTION AND UTTERANCE REJECTION which can be found here : https://arxiv.org/pdf/2201.03313.pdf

\n

few more related interesting works can be found here : https://github.com/facebookresearch/fairseq/blob/main/examples/nonautoregressive_translation/README.md

\n

maybe this discussion about spell correction can help you too : https://discuss.huggingface.co/t/asr-spell-correction/5103?page=2

", "rawMarkdown": "[NeuralSpeech](https://github.com/microsoft/NeuralSpeech) is a research project in Microsoft Research Asia focusing on neural network based speech processing, including automatic speech recognition (ASR), text to speech (TTS), etc.\n\nCurrently [this repo](https://github.com/microsoft/NeuralSpeech) covers several research work:\n\n**Automatic Speech Recognition**\n\n1. FastCorrect, NeurIPS 2021\n2. FastCorrect 2, EMNLP 2021 Findings\n3. CMatch for ASR, INTERSPEECH 2021\n4. Adapter for ASR, TASLP 2022\n\n**Text to Speech**\n5. LightSpeech, ICASSP 2021\n6. PriorGrad, ICLR 2022\n\nFor more research in NeuralSpeech project, you can refer to this page: https://speechresearch.github.io/. they will release more research work in the future.\n\nFastCorrect and FastCorrect 2 are worth exploring, i guess. FastCorrect paper link : https://arxiv.org/abs/2105.03842\n\nFastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition, by Yichong Leng, Xu Tan, Linchen Zhu, Jin Xu, Renqian Luo, Linquan Liu, Tao Qin, Xiang-Yang Li, Ed Lin and Tie-Yan Liu in NeurIPS 2021, is a novel non-autoregressive error correction method for automatic speech recognition. It use an encoder and a length predictor to detect errors and a decoder to correct errors.\n\n![](https://github.com/microsoft/NeuralSpeech/raw/master/img/FastCorrect_align.PNG)\n\nEdit alignment of FastCorrect\n![](https://github.com/microsoft/NeuralSpeech/raw/master/img/FastCorrect.PNG)\n\nThe whole pipeline of FastCorrect consists of 4 steps, you can find the details here : https://github.com/microsoft/NeuralSpeech/tree/master/FastCorrect\n\nanother interesting paper work is titled CROSS-MODAL ASR POST-PROCESSING SYSTEM FOR ERROR CORRECTION AND UTTERANCE REJECTION which can be found here : https://arxiv.org/pdf/2201.03313.pdf\n\nfew more related interesting works can be found here : https://github.com/facebookresearch/fairseq/blob/main/examples/nonautoregressive_translation/README.md\n\nmaybe this discussion about spell correction can help you too : https://discuss.huggingface.co/t/asr-spell-correction/5103?page=2\n\n", "votes": 7 } ], "comments": [], "raw_markdown_by_id": { "1845047": "[NeuralSpeech](https://github.com/microsoft/NeuralSpeech) is a research project in Microsoft Research Asia focusing on neural network based speech processing, including automatic speech recognition (ASR), text to speech (TTS), etc.\n\nCurrently [this repo](https://github.com/microsoft/NeuralSpeech) covers several research work:\n\n**Automatic Speech Recognition**\n\n1. FastCorrect, NeurIPS 2021\n2. FastCorrect 2, EMNLP 2021 Findings\n3. CMatch for ASR, INTERSPEECH 2021\n4. Adapter for ASR, TASLP 2022\n\n**Text to Speech**\n5. LightSpeech, ICASSP 2021\n6. PriorGrad, ICLR 2022\n\nFor more research in NeuralSpeech project, you can refer to this page: https://speechresearch.github.io/. they will release more research work in the future.\n\nFastCorrect and FastCorrect 2 are worth exploring, i guess. FastCorrect paper link : https://arxiv.org/abs/2105.03842\n\nFastCorrect: Fast Error Correction with Edit Alignment for Automatic Speech Recognition, by Yichong Leng, Xu Tan, Linchen Zhu, Jin Xu, Renqian Luo, Linquan Liu, Tao Qin, Xiang-Yang Li, Ed Lin and Tie-Yan Liu in NeurIPS 2021, is a novel non-autoregressive error correction method for automatic speech recognition. It use an encoder and a length predictor to detect errors and a decoder to correct errors.\n\n![](https://github.com/microsoft/NeuralSpeech/raw/master/img/FastCorrect_align.PNG)\n\nEdit alignment of FastCorrect\n![](https://github.com/microsoft/NeuralSpeech/raw/master/img/FastCorrect.PNG)\n\nThe whole pipeline of FastCorrect consists of 4 steps, you can find the details here : https://github.com/microsoft/NeuralSpeech/tree/master/FastCorrect\n\nanother interesting paper work is titled CROSS-MODAL ASR POST-PROCESSING SYSTEM FOR ERROR CORRECTION AND UTTERANCE REJECTION which can be found here : https://arxiv.org/pdf/2201.03313.pdf\n\nfew more related interesting works can be found here : https://github.com/facebookresearch/fairseq/blob/main/examples/nonautoregressive_translation/README.md\n\nmaybe this discussion about spell correction can help you too : https://discuss.huggingface.co/t/asr-spell-correction/5103?page=2\n\n" } }