{"id":29020214,"url":"https://github.com/deezer/robust-ai-lyrics-detection","last_synced_at":"2026-02-04T10:03:00.229Z","repository":{"id":295093633,"uuid":"989101534","full_name":"deezer/robust-AI-lyrics-detection","owner":"deezer","description":"Code to reproduce the experiments presented in the article \"Double Entendre: Robust Audio-Based AI-Generated Lyrics Detection via Multi-View Fusion\" (Findings of ACL 2025)","archived":false,"fork":false,"pushed_at":"2025-05-23T14:57:55.000Z","size":3,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2025-05-23T16:21:00.646Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":null,"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/deezer.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,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-05-23T14:57:25.000Z","updated_at":"2025-05-23T14:57:58.000Z","dependencies_parsed_at":"2025-05-23T16:21:03.861Z","dependency_job_id":"55f35e51-08d5-428e-939b-9c179bc769d6","html_url":"https://github.com/deezer/robust-AI-lyrics-detection","commit_stats":null,"previous_names":["deezer/robust-ai-lyrics-detection"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/deezer/robust-AI-lyrics-detection","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/deezer%2Frobust-AI-lyrics-detection","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/deezer%2Frobust-AI-lyrics-detection/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/deezer%2Frobust-AI-lyrics-detection/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/deezer%2Frobust-AI-lyrics-detection/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/deezer","download_url":"https://codeload.github.com/deezer/robust-AI-lyrics-detection/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/deezer%2Frobust-AI-lyrics-detection/sbom","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":261978911,"owners_count":23239417,"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":[],"created_at":"2025-06-26T01:04:32.787Z","updated_at":"2026-02-04T10:03:00.224Z","avatar_url":"https://github.com/deezer.png","language":null,"funding_links":[],"categories":[],"sub_categories":[],"readme":"# Double Entendre\n\n**Robust audio-based AI-generated lyrics detection via multi-view fusion.**\n\nCode to reproduce the experiments presented in *\"Double Entendre: Robust Audio-Based AI-Generated Lyrics Detection via Multi-View Fusion\"* (Findings of ACL 2025) and *\"AI-Generated Song Detection via Lyrics Transcripts\"* (ISMIR 2025).\n\nThis repository provides tools for detecting AI-generated music by analyzing sung lyrics extracted via speech transcription. It supports multiple transcribers, feature extractors (text and audio), and evaluation scenarios including audio perturbations and out-of-distribution generalization.\n\n## Overview\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"figures/dd_method.png\" alt=\"Double Entendre Method\" width=\"800\"\u003e\n\u003c/p\u003e\n\n**Supported scenarios:**\n- Real vs. Fake (AI-generated lyrics + AI vocals)\n- Real vs. Half-fake (human lyrics + AI vocals)\n- Cross-platform generalization (Suno → Udio)\n- Robustness under audio attacks (reverb, pitch, EQ, noise, time stretch)\n\n---\n\n## Project Structure\n\n```\n.\n├── transcription/       # Audio → text transcription\n│   ├── run_whisper.py   # Whisper transcription\n│   ├── run_meta.py      # Meta models (Seamless, MMS)\n│   └── ...\n├── features/            # Feature extraction \u0026 detection\n│   ├── feature_extractor.py  # Base class\n│   ├── sbert.py, llm2vec.py  # Text features\n│   ├── xeus.py, w2v2.py      # Speech features\n│   ├── run_features.py       # Main experiments\n│   └── ...\n├── pyproject.toml\n└── README.md\n```\n\nSee module READMEs for details:\n- [`transcription/README.md`](transcription/README.md)\n- [`features/README.md`](features/README.md)\n\n---\n\n## Installation\n\n```bash\ngit clone https://github.com/deezer/robust-AI-lyrics-detection.git\ncd robust-AI-lyrics-detection\npip install .\n```\n\n**Requirements:** Python 3.9–3.11, CUDA-capable GPU recommended.\n\n### Optional Dependencies\n\n```bash\npip install \".[meta]\"       # Meta speech models (Seamless, MMS)\npip install \".[demucs]\"     # Source separation (improves transcription)\npip install \".[xeus]\"       # XEUS speech features (requires checkpoint below)\npip install \".[binoculars]\" # Binoculars detector\n```\n\n### XEUS Checkpoint\n\nXEUS requires a manual checkpoint download:\n\n```bash\nwget https://huggingface.co/espnet/xeus/resolve/main/model/xeus_checkpoint.pth -P checkpoints/\n```\n\n---\n\n## Quick Start\n\n### 1. Transcribe\n\n```bash\ncd transcription\npython run_whisper.py --model large-v2 --file_path data/real/songs.json\n```\n\n### 2. Detect\n\n```bash\ncd features\npython run_features.py\n```\n\n---\n\n## Data Format\n\nDataset JSON files should follow this structure:\n\n```json\n{\n  \"train\": [\n    {\n      \"md5\": \"abc123...\",\n      \"text\": \"Original lyrics text...\",\n      \"language_str\": \"en\",\n      \"genre\": \"pop\",\n      \"artist_name\": \"Artist\",\n      \"class\": \"real\"\n    }\n  ],\n  \"test\": [...]\n}\n```\n\nFor AI-generated songs, include `mp3_url` or local `audio_path` instead of `md5`.\n\n| Field | Description |\n|-------|-------------|\n| `md5` | Audio file hash (for path lookup) |\n| `text` | Ground-truth lyrics |\n| `language_str` | ISO language code |\n| `class` | `real`, `generated`, or `half-fake` |\n| `transcription` | Added by transcription scripts |\n\n---\n\n## Configuration\n\nEnvironment variables for path configuration:\n\n```bash\nexport AUDIO_BASE_PATH=/path/to/audio      # Audio file storage\nexport PROJECT_BASE_DIR=/path/to/project   # Project root\nexport ARTEFACTS_DIR=./artefacts           # Model outputs\n```\n\n---\n\n## Citation\n\nIf you use this code, please cite:\n\n### Double Entendre (Findings of ACL 2025)\n\n```bibtex\n@inproceedings{frohmann2025double,\n    author    = {Frohmann, Markus and Meseguer-Brocal, Gabriel and \n                 Epure, Elena V. and Labrak, Yanis},\n    title     = {{Double Entendre}: Robust Audio-Based {AI}-Generated Lyrics \n                 Detection via Multi-View Fusion},\n    booktitle = {Findings of the Association for Computational Linguistics: ACL 2025},\n    year      = {2025},\n    publisher = {Association for Computational Linguistics}\n}\n```\n\n### AI-Generated Song Detection via Lyrics Transcripts (ISMIR 2025)\n\n```bibtex\n@inproceedings{frohmann2025ai,\n    author    = {Frohmann, Markus and Epure, Elena and \n                 Meseguer Brocal, Gabriel and Schedl, Markus and Hennequin, Romain},\n    title     = {{AI}-Generated Song Detection via Lyrics Transcripts},\n    booktitle = {Proceedings of the 26th International Society for \n                 Music Information Retrieval Conference (ISMIR)},\n    year      = {2025},\n    pages     = {121--130},\n    address   = {Daejeon, South Korea},\n    doi       = {10.5281/zenodo.17706345}\n}\n```\n\n---\n\n## Related Work\n\n- [synthetic_lyrics_detection](https://github.com/deezer/synthetic_lyrics_detection) — Synthetic lyrics generation pipeline (TrustNLP 2025)\n\n---\n\n## License\n\nMIT License. See [LICENSE](LICENSE).\n\nBuilt at [Deezer Research](https://research.deezer.com/), Paris.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdeezer%2Frobust-ai-lyrics-detection","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fdeezer%2Frobust-ai-lyrics-detection","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fdeezer%2Frobust-ai-lyrics-detection/lists"}