{
 "id": "ModelsLab/midashenglm-gen-training-latents",
 "owner": "ModelsLab",
 "name": "midashenglm-gen-training-latents",
 "downloads": 144,
 "likes": 0,
 "last_modified": "2026-08-23T18:44:31.000Z",
 "stated_licence": "cc-by-4.0",
 "stated_from": "tag",
 "licence_name": null,
 "licence_link": null,
 "file_licence": null,
 "licence_files": [],
 "sources": [],
 "gated": false,
 "tasks": [
  "text-to-audio"
 ],
 "citation": "Released CC-BY-4.0, and derived from two attribution-only sources that must be\ncredited by anyone using it:\n\n- **LibriTTS-R** ([`mythicinfinity/libritts_r`](https://huggingface.co/datasets/mythicinfinity/libritts_r)),\n  CC-BY-4.0 — Koizumi et al., *LibriTTS-R: A Restored Multi-Speaker Text-to-Speech Corpus*.\n- **MECAT-Caption** ([`mispeech/MECAT-Caption`](https://huggingface.co/datasets/mispeech/MECAT-Caption)),\n  CC-BY-3.0 — the public slice of ACAVCaps, itself derived from ACAV100M.\n\nBecause the model's decoder reconstructs audio from these latents at parity with\nthe source — log-mel L1 of 0.26 against controls at 3.3-4.0, and identical word\nerror rate — treat this as a faithful encoding of that audio rather than as a\nderived feature set, and honour the source licences accordingly.\n\nBase model `mispeech/midashenglm-gen` is Apache 2.0.",
 "signal": "permissive",
 "reason": "The declared licence (cc-by-4.0) is classified as permissive.",
 "revision": "1435234a33bd90f514a76dc079e00c9786439fe3",
 "stated_evidence": "https://huggingface.co/datasets/ModelsLab/midashenglm-gen-training-latents/blob/1435234a33bd90f514a76dc079e00c9786439fe3/README.md",
 "file_evidence": "https://huggingface.co/datasets/ModelsLab/midashenglm-gen-training-latents/tree/1435234a33bd90f514a76dc079e00c9786439fe3",
 "card_nc": false
}