{"id":42034085,"url":"https://github.com/HawkAaron/RNN-Transducer","last_synced_at":"2026-02-05T04:01:09.764Z","repository":{"id":49967416,"uuid":"128898651","full_name":"HawkAaron/RNN-Transducer","owner":"HawkAaron","description":"MXNet implementation of RNN Transducer (Graves 2012): Sequence Transduction with Recurrent Neural Networks","archived":false,"fork":false,"pushed_at":"2021-06-07T15:39:34.000Z","size":50,"stargazers_count":139,"open_issues_count":10,"forks_count":31,"subscribers_count":6,"default_branch":"graves2013","last_synced_at":"2026-02-01T21:27:22.614Z","etag":null,"topics":["asr","end-to-end","mxnet","rnn-transducer","rnnt-joint","rnnt-model","sequence-transduction","speech-recognition","timit","transducers"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/HawkAaron.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null}},"created_at":"2018-04-10T08:20:19.000Z","updated_at":"2025-03-09T09:37:56.000Z","dependencies_parsed_at":"2022-09-07T07:40:14.149Z","dependency_job_id":null,"html_url":"https://github.com/HawkAaron/RNN-Transducer","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/HawkAaron/RNN-Transducer","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/HawkAaron%2FRNN-Transducer","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/HawkAaron%2FRNN-Transducer/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/HawkAaron%2FRNN-Transducer/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/HawkAaron%2FRNN-Transducer/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/HawkAaron","download_url":"https://codeload.github.com/HawkAaron/RNN-Transducer/tar.gz/refs/heads/graves2013","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/HawkAaron%2FRNN-Transducer/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":29110544,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-02-05T03:44:17.043Z","status":"ssl_error","status_checked_at":"2026-02-05T03:44:12.077Z","response_time":65,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.5:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["asr","end-to-end","mxnet","rnn-transducer","rnnt-joint","rnnt-model","sequence-transduction","speech-recognition","timit","transducers"],"created_at":"2026-01-26T05:00:43.188Z","updated_at":"2026-02-05T04:01:09.725Z","avatar_url":"https://github.com/HawkAaron.png","language":"Python","funding_links":[],"categories":["\u003ca name=\"NLP\"\u003e\u003c/a\u003e3. NLP"],"sub_categories":[],"readme":"# End-to-End Speech Recognition using RNN-Transducer\n## File description\n* eval.py: rnnt joint model decode\n* model.py: rnnt model, which contains acoustic / phoneme model\n* model2012.py: rnnt model refer to Graves2012\n* seq2seq/*: seq2seq with attention \n* rnnt_np.py: rnnt loss function implementation on mxnet, support for both symbol and gluon [refer to PyTorch implementation](https://github.com/awni/transducer)\n* DataLoader.py: data process\n* train.py: rnnt training script, can be initialized from CTC and PM model\n* train_ctc.py: ctc training script\n* train_att.py: attention training script\n\n## Directory description\n* conf: kaldi feature extraction config\n\n## Reference Paper\n* RNN Transducer (Graves 2012): [Sequence Transduction with Recurrent Neural Networks](https://arxiv.org/abs/1211.3711)\n* RNNT joint (Graves 2013): [Speech Recognition with Deep Recurrent Neural Networks](https://arxiv.org/abs/1303.5778 )\n* E2E criterion comparison (Baidu 2017): [Exploring Neural Transducers for End-to-End Speech Recognition](https://arxiv.org/abs/1707.07413)\n* Seq2Seq-Attention: [Attention-Based Models for Speech Recognition](https://arxiv.org/abs/1506.07503)\n\n## Run\n* Compile RNNT Loss\nFollow the instructions in [here](https://github.com/HawkAaron/mxnet-transducer/tree/master) to compile MXNET with RNNT loss.\n\n* Extract feature\nlink kaldi timit example dirs (`local` `steps` `utils` )\nexcute `run.sh` to extract 40 dim fbank feature\nrun `feature_transform.sh` to get 123 dim feature as described in Graves2013\n\n* Train RNNT model:\n```bash\npython train.py --lr 1e-3 --bi --dropout .5 --out exp/rnnt_bi_lr1e-3 --schedule\n```\n\n## Evaluation\nDefault only for RNNT\n* Greedy decoding:\n```\npython eval.py \u003cpath to best model parameters\u003e --bi\n```\n* Beam search:\n```\npython eval.py \u003cpath to best model parameters\u003e --bi --beam \u003cbeam size\u003e\n```\n\n## Results\n* CTC \n\n    | Decode | PER |\n    | --- | --- |\n    | greedy | 20.36 |\n    | beam 100 | 20.03 |\n\n* Transducer\n\n    | Decode | PER |\n    | --- | --- |\n    | greedy | 20.74 |\n    | beam 40 | 19.84 |\n\n## Requirements\n* Python 3.6\n* MxNet 1.1.0\n* numpy 1.14\n\n## TODO\n* beam serach accelaration\n* Seq2Seq with attention\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FHawkAaron%2FRNN-Transducer","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FHawkAaron%2FRNN-Transducer","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FHawkAaron%2FRNN-Transducer/lists"}