ggfox00000/dia-AMICorpus-all

--- license: cc-by-4.0 task_categories: - automatic-speech-recognition - voice-activity-detection language: - en size_categories: - n<1K pretty_name: AMI Meeting Corpus — full mirror (Mix-Headset + manual annotations) tags: - ami - meeting - speaker-diarization - diarization - asr - rttm - dialogue-acts - nxt - audio annotations_creators: - expert-generated source_datasets: - extended|ami-meeting-corpus dataset_info: features: - name: audio dtype: audio - name: file_id dtype: string splits: - name: all num_examples: 171 --- # AMI Meeting Corpus — Full Mirror (Mix-Headset + manual annotations) Miroir **complet** du AMI Meeting Corpus distribué par l'AMI Consortium (Edinburgh). Tous les fichiers sont repris *tels quels* depuis la distribution upstream, y compris la structure de dossiers. ## Contenu - **171 meetings** (~100 h d'audio) — scénarios scénarisés (ES/IS/TS) et réunions naturelles (EN/IB/IN) - **Audio Mix-Headset** : `amicorpus/<meeting>/audio/<meeting>.Mix-Headset.wav` — mixdown des micros serre-tête (un par speaker), c'est le flux audio standard pour benchmark diarisation / ASR close-talk - **Manual annotations v1.6.2** (NXT XML) dans `ami_public_manual_1.6.2/` : segments, words, dialogueActs, topics, abstractive/extractive summaries, namedEntities, focus, gestures (hand/head), movement, disfluency, argumentation, participantRoles, participantSummaries, decision, youUsages, ontologies, corpusResources… - Langue : **anglais** (UK/EU — meetings AMI) - Licence : **CC-BY-4.0** (cf `CCBY4.0.txt` à la racine) ## Structure ``` dia-AMICorpus-all/ ├── README.md ├── CCBY4.0.txt ├── amiBuild-*.manifest.txt ├── amicorpus/ │ └── <meeting>/audio/<meeting>.Mix-Headset.wav # 171 WAV └── ami_public_manual_1.6.2/ # NXT v1.6.2 ├── 00README_MANUAL.txt ├── LICENCE.txt ├── AMI-metadata.xml ├── MANIFEST_MANUAL.txt ├── manifest_1.7.html ├── resource.xml ├── abstractive/ # 142 fichiers ├── argumentation/ # 3 ├── configuration/ # 2 ├── corpusResources/ # 2 ├── corpusdoc/ # 73 ├── decision/ # 1 ├── dialogueActs/ # 695 ├── disfluency/ # 160 ├── extractive/ # 274 ├── focus/ # 56 ├── handGesture/ # 61 ├── headGesture/ # 173 ├── movement/ # 498 ├── namedEntities/ # 468 ├── ontologies/ # 17 ├── participantRoles/ # 20 ├── participantSummaries/ # 323 ├── segments/ # 687 ├── topics/ # 139 ├── words/ # 687 └── youUsages/ # 63 ``` ## Meetings disponibles (171) - Scénarios (ES/IS/TS) : ES2002–ES2016, IS1000–IS1009, TS3003–TS3012 - Naturels (EN) : EN2001–EN2009 - Instrumental / natural (IB/IN) : IB4001–IB4011, IN1001–IN1016 ## Format NXT (exemple segments) ```xml <?xml version="1.0" encoding="ISO-8859-1"?> <nite:root nite:id="ES2002a.A.segs" xmlns:nite="http://nite.sourceforge.net/"> <segment nite:id="ES2002a.sync.1" transcriber_start="12.34" transcriber_end="14.56" participant="A"> <nite:child href="ES2002a.A.words.xml#id(ES2002a.w.12)..id(ES2002a.w.27)"/> </segment> ... </nite:root> ``` Pour construire un RTTM standard à partir des segments NXT : ```python from glob import glob from xml.etree import ElementTree as ET NS = {"nite": "http://nite.sourceforge.net/"} for seg_xml in glob("ami_public_manual_1.6.2/segments/*.segments.xml"): root = ET.parse(seg_xml).getroot() meeting, spk, _ = seg_xml.split("/")[-1].split(".", 2) for seg in root.findall("segment"): t0 = float(seg.attrib.get("transcriber_start", seg.attrib.get("starttime", 0))) t1 = float(seg.attrib.get("transcriber_end", seg.attrib.get("endtime", 0))) if t1 > t0: print(f"SPEAKER {meeting} 1 {t0:.3f} {t1-t0:.3f} <NA> <NA> {spk} <NA> <NA>") ``` ## Utilisation ```python from huggingface_hub import snapshot_download root = snapshot_download("ggfox00000/dia-AMICorpus-all", repo_type="dataset") # root/amicorpus/<meeting>/audio/<meeting>.Mix-Headset.wav # root/ami_public_manual_1.6.2/segments/<meeting>.<spk>.segments.xml # root/ami_public_manual_1.6.2/words/<meeting>.<spk>.words.xml ``` ## Source - AMI Meeting Corpus — https://groups.inf.ed.ac.uk/ami/corpus/ - Annotations NXT v1.6.2 — AMI Consortium - Carletta et al. 2005, *The AMI Meeting Corpus: A Pre-Announcement*, MLMI ## Licence **CC-BY-4.0** — voir `CCBY4.0.txt`. Distribution miroir de la distribution officielle du AMI Consortium pour faciliter l'accès en recherche. ## Citation ```bibtex @inproceedings{carletta2005ami, author = {Carletta, Jean and Ashby, Simone and Bourban, Sebastien and Flynn, Mike and Guillemot, Mael and Hain, Thomas and Kadlec, Jaroslav and Karaiskos, Vasilis and Kraaij, Wessel and Kronenthal, Melissa and Lathoud, Guillaume and Lincoln, Mike and Lisowska, Agnes and McCowan, Iain and Post, Wilfried and Reidsma, Dennis and Wellner, Pierre}, title = {The {AMI} Meeting Corpus: A Pre-Announcement}, booktitle = {MLMI}, year = {2005}, pages = {28--39}, } ```--- 许可证：CC-BY-4.0 任务类别： - 自动语音识别（automatic-speech-recognition） - 语音活动检测（voice-activity-detection） 语言： - 英语（en） 样本规模： - 少于1000条（n<1K） 正式名称：AMI会议语料库完整镜像版（混合头戴式麦克风音频+人工标注） 标签： - ami - 会议 - 说话人日志（speaker-diarization） - 日志（diarization） - asr - rttm - 对话行为（dialogue-acts） - nxt - 音频 标注生成方： - 专家生成（expert-generated） 源数据集： - 扩展源|ami-meeting-corpus 数据集信息： 特征字段： - 名称：audio，数据类型：音频类型 - 名称：file_id，数据类型：字符串 数据集划分： - 划分名称：all（全部），样本数量：171 --- # AMI会议语料库完整镜像版（混合头戴式麦克风音频+人工标注） 由爱丁堡AMI联盟发布的AMI会议语料库完整镜像版本。所有文件均直接取自上游发布版本，完整保留原目录结构。 ## 数据集内容 - **171场会议**：总时长约100小时，涵盖脚本式会议（ES/IS/TS系列）与自然会话会议（EN/IB/IN系列） - **混合头戴式麦克风音频**：路径为`amicorpus/<会议ID>/audio/<会议ID>.Mix-Headset.wav`，该音频为每位说话者头戴式麦克风的混音结果，是用于说话人日志（speaker diarization）与近距离自动语音识别基准测试的标准音频流。 - **v1.6.2版人工标注（NXT XML格式）**：存放于`ami_public_manual_1.6.2/`目录下，包含片段、单词、对话行为、主题、抽象/抽取式摘要、命名实体、注意力焦点、手势（手部/头部）、动作、言语障碍、论证结构、参与者角色、参与者摘要、决策记录、you指代使用、本体、语料库资源等标注内容。 - **语言**：英语（英国/欧盟口音，适配AMI会议场景） - **许可证**：CC-BY-4.0（详见根目录下的`CCBY4.0.txt`文件） ## 目录结构 dia-AMICorpus-all/ ├── README.md ├── CCBY4.0.txt ├── amiBuild-*.manifest.txt ├── amicorpus/ │ └── <meeting>/audio/<meeting>.Mix-Headset.wav # 171个WAV音频文件 └── ami_public_manual_1.6.2/ # NXT v1.6.2标注集 ├── 00README_MANUAL.txt ├── LICENCE.txt ├── AMI-metadata.xml ├── MANIFEST_MANUAL.txt ├── manifest_1.7.html ├── resource.xml ├── abstractive/ # 142个文件 ├── argumentation/ # 3个文件 ├── configuration/ # 2个文件 ├── corpusResources/ # 2个文件 ├── corpusdoc/ # 73个文件 ├── decision/ # 1个文件 ├── dialogueActs/ # 695个文件 ├── disfluency/ # 160个文件 ├── extractive/ # 274个文件 ├── focus/ # 56个文件 ├── handGesture/ # 61个文件 ├── headGesture/ # 173个文件 ├── movement/ # 498个文件 ├── namedEntities/ # 468个文件 ├── ontologies/ # 17个文件 ├── participantRoles/ # 20个文件 ├── participantSummaries/ # 323个文件 ├── segments/ # 687个文件 ├── topics/ # 139个文件 ├── words/ # 687个文件 └── youUsages/ # 63个文件 ## 可用会议列表（共171场） - 脚本式会议（ES/IS/TS系列）：ES2002–ES2016、IS1000–IS1009、TS3003–TS3012 - 自然会话会议（EN系列）：EN2001–EN2009 - 半自然/自然混合会议（IB/IN系列）：IB4001–IB4011、IN1001–IN1016 ## NXT格式示例（以片段标注为例） xml <?xml version="1.0" encoding="ISO-8859-1"?> <nite:root nite:id="ES2002a.A.segs" xmlns:nite="http://nite.sourceforge.net/"> <segment nite:id="ES2002a.sync.1" transcriber_start="12.34" transcriber_end="14.56" participant="A"> <nite:child href="ES2002a.A.words.xml#id(ES2002a.w.12)..id(ES2002a.w.27)"/> </segment> ... </nite:root> 以下为从NXT片段标注生成标准RTTM（Rich Transcription Time Marked）格式文件的Python示例代码： python from glob import glob from xml.etree import ElementTree as ET NS = {"nite": "http://nite.sourceforge.net/"} for seg_xml in glob("ami_public_manual_1.6.2/segments/*.segments.xml"): root = ET.parse(seg_xml).getroot() meeting, spk, _ = seg_xml.split("/")[-1].split(".", 2) for seg in root.findall("segment"): t0 = float(seg.attrib.get("transcriber_start", seg.attrib.get("starttime", 0))) t1 = float(seg.attrib.get("transcriber_end", seg.attrib.get("endtime", 0))) if t1 > t0: print(f"SPEAKER {meeting} 1 {t0:.3f} {t1-t0:.3f} <NA> <NA> {spk} <NA> <NA>") ## 使用方法 python from huggingface_hub import snapshot_download root = snapshot_download("ggfox00000/dia-AMICorpus-all", repo_type="dataset") # root/amicorpus/<meeting>/audio/<meeting>.Mix-Headset.wav # root/ami_public_manual_1.6.2/segments/<meeting>.<spk>.segments.xml # root/ami_public_manual_1.6.2/words/<meeting>.<spk>.words.xml ## 数据集来源 - AMI会议语料库：https://groups.inf.ed.ac.uk/ami/corpus/ - NXT v1.6.2版标注：AMI联盟 - 引用文献：Carletta等人，2005年，《AMI会议语料库：预发布公告》，发表于MLMI会议 ## 许可证 **CC-BY-4.0**，详见根目录下的`CCBY4.0.txt`文件。本镜像为AMI联盟官方发布版本的完整副本，旨在便于科研人员获取。 ## 引用格式 bibtex @inproceedings{carletta2005ami, author = {Carletta, Jean and Ashby, Simone and Bourban, Sebastien and Flynn, Mike and Guillemot, Mael and Hain, Thomas and Kadlec, Jaroslav and Karaiskos, Vasilis and Kraaij, Wessel and Kronenthal, Melissa and Lathoud, Guillaume and Lincoln, Mike and Lisowska, Agnes and McCowan, Iain and Post, Wilfried and Reidsma, Dennis and Wellner, Pierre}, title = {The {AMI} Meeting Corpus: A Pre-Announcement}, booktitle = {MLMI}, year = {2005}, pages = {28--39}, }