{"id":13861957,"url":"https://github.com/aisu-programming/Chord-Recognition","last_synced_at":"2025-07-14T11:32:04.814Z","repository":{"id":43179434,"uuid":"306082279","full_name":"aisu-programming/Chord-Recognition","owner":"aisu-programming","description":"Recognize chords in songs using Bidirectional Transformer || AI Cup 2020 - Chord Recognition Competition (9th place) / 和弦辨識競賽 (第九名)","archived":false,"fork":false,"pushed_at":"2024-04-04T01:21:12.000Z","size":258314,"stargazers_count":15,"open_issues_count":0,"forks_count":5,"subscribers_count":2,"default_branch":"main","last_synced_at":"2024-08-05T06:04:32.029Z","etag":null,"topics":["chord-estimation","chord-recognition","deep-learning","tensorflow","transformer"],"latest_commit_sha":null,"homepage":"https://aidea-web.tw/topic/43d9cc47-b70e-4751-80d3-a2d7333eb77b","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/aisu-programming.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2020-10-21T16:21:28.000Z","updated_at":"2024-07-23T18:38:47.000Z","dependencies_parsed_at":"2024-08-05T06:14:09.647Z","dependency_job_id":null,"html_url":"https://github.com/aisu-programming/Chord-Recognition","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aisu-programming%2FChord-Recognition","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aisu-programming%2FChord-Recognition/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aisu-programming%2FChord-Recognition/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/aisu-programming%2FChord-Recognition/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/aisu-programming","download_url":"https://codeload.github.com/aisu-programming/Chord-Recognition/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":225974255,"owners_count":17553913,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","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":["chord-estimation","chord-recognition","deep-learning","tensorflow","transformer"],"created_at":"2024-08-05T06:01:33.717Z","updated_at":"2024-11-22T22:30:21.154Z","avatar_url":"https://github.com/aisu-programming.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"# Chord-Recognition\n---\n[中文版 README](README_zh-TW.md)\n\n[AI CUP 2020 - Chord Recognition Competition Website](https://aidea-web.tw/topic/43d9cc47-b70e-4751-80d3-a2d7333eb77b)\n\nAwards: https://global.turingcerts.com/en/co/cert?hash=6a1e9c33453834e3eb8f201e558c563868a308f430606bddb863fe89b6638171\n\n---\n\n## Environment\n- Hardware：\n  - CPU：i7-9700\n  - GPU：RTX 2070\n- Software：\n  - Cuda：cuda_**11.0**.2_451.48_win10\n  - cuDNN：cudnn-11.0-windows-x64-v**8.0.4**.30\n  - TensorFlow：tensorflow-**2.4**.0\n\n## Development Flow\n1. `answerAnalyze.py`:\n\n   Observe training data through CMD.\n   \n2. `visualization.py`:\n\n   Observe training data much more conveniently by matplotlib.\n   \n3. `score.py`:\n   \n   Comprehend the principle of scoring method \u0026 implement the scoring program.\n   \n4. `main.py` → `processData.py`: \n   \n   Preprocess the training data into the input data for models.\n   \n5. `model.py`: \n   \n   Start building the fisrt version model.\n   \n6. `mapping.py`: \n   \n   Including all required mapping dictionaries of input data Y for the first version model.\n   These mapping dictionaries concludes 544 possible input data Ys in training data -- \"CE200\".\n   \n7. `model.py` → `trainModel.py` → `oneFrameModel.py`: \n   \n   Modularize models \u0026 rename it to one-frame-input predicting model.\n   \n8. `multiFrameModel.py`: \n   \n   Start building the second version model, which can use multiple frames as input data X.\n   \n9. Improve `processData.py`:\n   \n   Divide input data X during preprocessing, which significantly improved the accuracy.\n\n\u003e p.s.\n\u003e So far, I always input \"random 40% part of one song\" to predict \"answers of the rest 60% part of the song\".\n\u003e The oneFrameModel's accuracy is about 80%, and the multiFrameModel can achieve 99.9%.\n\u003e I then realized that this is an incorrect predicting method in the next step.\n\n10. `multiFrameModel-2.py` → `splitDataModel.py`:\n\n    Extends the conception of multiple-frame input \u0026 dividing input data X.\n    Changing predicting pattern from \"60% for a song -\u003e answers of the rest 40%\" to \"60% songs -\u003e the rest 40% songs\".\n    Ex: There are 20 songs in CE200_sample, so I will choose 8 songs as training data, and the rest 12 songs are for validation data. \n    After changing, the accuracy drops to 45%.\n    \n11. Improve `mapping.py`:\n    \n    Adopt new mapping dictionary which only contains scoring data Ys.\n    Accuracy raise from 45% to 50%.\n    \n12. `fasterReadingModel.py`：\n    \n    Adjust the time of preprocessing to the instant before inputting data X.\n    This significantly reduce the time for data reading \u0026 the GPU's RAM usage.\n    \n## Result\n- Got the 9th place.\n- The 1st place used the model in [this paper](https://paperswithcode.com/paper/feature-learning-for-chord-recognition-the).\n- The second place used the same model with me, but more preprocess \u0026 Post-processing.\n- Review: It seems like that information in data are much more important than the architecture of the model. Maybe I should do feature extraction on my own in the next time.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Faisu-programming%2FChord-Recognition","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Faisu-programming%2FChord-Recognition","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Faisu-programming%2FChord-Recognition/lists"}