add-stream

README.md

@@ -1,13 +1,74 @@
-# AMI Corpus
-
-https://groups.inf.ed.ac.uk/ami/corpus/
-
-To be filled!
+---
+annotations_creators: []
+language:
+- en
+language_creators: []
+license:
+- cc-by-4.0
+multilinguality:
+- monolingual
+pretty_name: AMI
+size_categories: []
+source_datasets: []
+tags: []
+task_categories:
+- automatic-speech-recognition
+task_ids: []
+---
+
+# Dataset Card for AMI
+
+## Table of Contents
+- [Table of Contents](#table-of-contents)
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages](#languages)
+- [Dataset Structure](#dataset-structure)
+  - [Data Instances](#data-instances)
+  - [Data Fields](#data-fields)
+  - [Data Splits](#data-splits)
+- [Dataset Creation](#dataset-creation)
+  - [Curation Rationale](#curation-rationale)
+  - [Source Data](#source-data)
+  - [Annotations](#annotations)
+  - [Personal and Sensitive Information](#personal-and-sensitive-information)
+- [Considerations for Using the Data](#considerations-for-using-the-data)
+  - [Social Impact of Dataset](#social-impact-of-dataset)
+  - [Discussion of Biases](#discussion-of-biases)
+  - [Other Known Limitations](#other-known-limitations)
+- [Additional Information](#additional-information)
+  - [Dataset Curators](#dataset-curators)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+  - [Contributions](#contributions)
+- [Terms of Usage](#terms-of-usage)
+  
+
+## Dataset Description
+
+- **Homepage:** https://groups.inf.ed.ac.uk/ami/corpus/
+- **Repository:** https://github.com/kaldi-asr/kaldi/tree/master/egs/ami/s5 
+- **Paper:** 
+- **Leaderboard:** 
+- **Point of Contact:** [[email protected]](mailto:[email protected])
+
+## Dataset Description
+
+The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
+synchronized to a common timeline. These include close-talking and far-field microphones, individual and
+room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
+the participants also have unsynchronized pens available to them that record what is written. The meetings
+were recorded in English using three different rooms with different acoustic properties, and include mostly
+non-native speakers.
 
 **Note**: This dataset corresponds to the data-processing of [KALDI's AMI S5 recipe](https://github.com/kaldi-asr/kaldi/tree/master/egs/ami/s5).
 This means text is normalized and the audio data is chunked according to the scripts above!
 To make the user experience as simply as possible, we provide the already chunked data to the user here so that the following can be done:
 
+
+### Example Usage
+
 ```python
 from datasets import load_dataset
 ds = load_dataset("edinburghcstr/ami", "ihm")
@@ -18,15 +79,15 @@ gives:
 ```
 DatasetDict({
     train: Dataset({
-        features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
+        features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
         num_rows: 108502
     })
     validation: Dataset({
-        features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
+        features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
         num_rows: 13098
     })
     test: Dataset({
-        features: ['segment_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
+        features: ['meeting_id', 'audio_id', 'text', 'audio', 'begin_time', 'end_time', 'microphone_id', 'speaker_id'],
         num_rows: 12643
     })
 })
@@ -39,10 +100,10 @@ ds["train"][0]
 automatically loads the audio into memory:
 
 ```
-{'segment_id': 'EN2001a',
+{'meeting_id': 'EN2001a',
  'audio_id': 'AMI_EN2001a_H00_MEE068_0000557_0000594',
  'text': 'OKAY',
- 'audio': {'path': '/home/patrick_huggingface_co/.cache/huggingface/datasets/downloads/extracted/2d75d5b3e8a91f44692e2973f08b4cac53698f92c2567bd43b41d19c313a5280/EN2001a/train_ami_en2001a_h00_mee068_0000557_0000594.wav',
+ 'audio': {'path': '/cache/dir/path/downloads/extracted/2d75d5b3e8a91f44692e2973f08b4cac53698f92c2567bd43b41d19c313a5280/EN2001a/train_ami_en2001a_h00_mee068_0000557_0000594.wav',
   'array': array([0.        , 0.        , 0.        , ..., 0.00033569, 0.00030518,
          0.00030518], dtype=float32),
   'sampling_rate': 16000},
@@ -70,4 +131,66 @@ The results are in-line with results of published papers:
 - [*Hybrid acoustic models for distant and multichannel large vocabulary speech recognition*](https://www.researchgate.net/publication/258075865_Hybrid_acoustic_models_for_distant_and_multichannel_large_vocabulary_speech_recognition)
 - [Multi-Span Acoustic Modelling using Raw Waveform Signals](https://arxiv.org/abs/1906.11047)
 
-You can run [run.sh](https://huggingface.co/patrickvonplaten/ami-wav2vec2-large-lv60/blob/main/run.sh) to reproduce the result.
+You can run [run.sh](https://huggingface.co/patrickvonplaten/ami-wav2vec2-large-lv60/blob/main/run.sh) to reproduce the result.
+
+### Supported Tasks and Leaderboards
+
+### Languages
+
+## Dataset Structure
+
+### Data Instances
+
+### Data Fields
+
+### Data Splits
+
+#### Transcribed Subsets Size
+
+## Dataset Creation
+
+### Curation Rationale
+
+### Source Data
+
+#### Initial Data Collection and Normalization
+
+#### Who are the source language producers?
+
+### Annotations
+
+#### Annotation process
+
+#### Who are the annotators?
+
+### Personal and Sensitive Information
+
+## Considerations for Using the Data
+
+### Social Impact of Dataset
+
+[More Information Needed]
+
+### Discussion of Biases
+
+### Other Known Limitations
+
+## Additional Information
+
+### Dataset Curators
+
+
+### Licensing Information
+
+
+### Citation Information
+
+
+### Contributions
+
+Thanks to [@sanchit-gandhi](https://github.com/sanchit-gandhi), [@patrickvonplaten](https://github.com/patrickvonplaten), 
+and [@polinaeterna](https://github.com/polinaeterna) for adding this dataset.
+
+## Terms of Usage
+
+

ami.py

@@ -1,4 +1,4 @@
-# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
+# Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -12,71 +12,57 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """
-GigaSpeech is an evolving, multi-domain English speech recognition corpus with 10,000 hours of high quality
-labeled audio suitable for supervised training, and 40,000 hours of total audio suitable for semi-supervised
-and unsupervised training. Around 40,000 hours of transcribed audio is first collected from audiobooks, podcasts
-and YouTube, covering both read and spontaneous speaking styles, and a variety of topics, such as arts, science,
-sports, etc. A new forced alignment and segmentation pipeline is proposed to create sentence segments suitable
-for speech recognition training, and to filter out segments with low-quality transcription. For system training,
-GigaSpeech provides five subsets of different sizes, 10h, 250h, 1000h, 2500h, and 10000h.
-For our 10,000-hour XL training subset, we cap the word error rate at 4% during the filtering/validation stage,
-and for all our other smaller training subsets, we cap it at 0%. The DEV and TEST evaluation sets, on the other hand,
-are re-processed by professional human transcribers to ensure high transcription quality.
+The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
+synchronized to a common timeline. These include close-talking and far-field microphones, individual and
+room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
+the participants also have unsynchronized pens available to them that record what is written. The meetings
+were recorded in English using three different rooms with different acoustic properties, and include mostly
+non-native speakers.
 """
 
-import csv
 import os
 
 import datasets
 
 _CITATION = """\
-@article{DBLP:journals/corr/abs-2106-06909,
-  author    = {Guoguo Chen and
-               Shuzhou Chai and
-               Guanbo Wang and
-               Jiayu Du and
-               Wei{-}Qiang Zhang and
-               Chao Weng and
-               Dan Su and
-               Daniel Povey and
-               Jan Trmal and
-               Junbo Zhang and
-               Mingjie Jin and
-               Sanjeev Khudanpur and
-               Shinji Watanabe and
-               Shuaijiang Zhao and
-               Wei Zou and
-               Xiangang Li and
-               Xuchen Yao and
-               Yongqing Wang and
-               Yujun Wang and
-               Zhao You and
-               Zhiyong Yan},
-  title     = {GigaSpeech: An Evolving, Multi-domain {ASR} Corpus with 10, 000 Hours
-               of Transcribed Audio},
-  journal   = {CoRR},
-  volume    = {abs/2106.06909},
-  year      = {2021},
-  url       = {https://arxiv.org/abs/2106.06909},
-  eprinttype = {arXiv},
-  eprint    = {2106.06909},
-  timestamp = {Wed, 29 Dec 2021 14:29:26 +0100},
-  biburl    = {https://dblp.org/rec/journals/corr/abs-2106-06909.bib},
-  bibsource = {dblp computer science bibliography, https://dblp.org}
+@inproceedings{10.1007/11677482_3,
+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},
+year = {2005},
+isbn = {3540325492},
+publisher = {Springer-Verlag},
+address = {Berlin, Heidelberg},
+url = {https://doi.org/10.1007/11677482_3},
+doi = {10.1007/11677482_3},
+abstract = {The AMI Meeting Corpus is a multi-modal data set consisting of 100 hours of meeting
+recordings. It is being created in the context of a project that is developing meeting
+browsing technology and will eventually be released publicly. Some of the meetings
+it contains are naturally occurring, and some are elicited, particularly using a scenario
+in which the participants play different roles in a design team, taking a design project
+from kick-off to completion over the course of a day. The corpus is being recorded
+using a wide range of devices including close-talking and far-field microphones, individual
+and room-view video cameras, projection, a whiteboard, and individual pens, all of
+which produce output signals that are synchronized with each other. It is also being
+hand-annotated for many different phenomena, including orthographic transcription,
+discourse properties such as named entities and dialogue acts, summaries, emotions,
+and some head and hand gestures. We describe the data set, including the rationale
+behind using elicited material, and explain how the material is being recorded, transcribed
+and annotated.},
+booktitle = {Proceedings of the Second International Conference on Machine Learning for Multimodal Interaction},
+pages = {28–39},
+numpages = {12},
+location = {Edinburgh, UK},
+series = {MLMI'05}
 }
 """
 
 _DESCRIPTION = """\
-GigaSpeech is an evolving, multi-domain English speech recognition corpus with 10,000 hours of high quality
-labeled audio suitable for supervised training, and 40,000 hours of total audio suitable for semi-supervised
-and unsupervised training. Around 40,000 hours of transcribed audio is first collected from audiobooks, podcasts
-and YouTube, covering both read and spontaneous speaking styles, and a variety of topics, such as arts, science,
-sports, etc. A new forced alignment and segmentation pipeline is proposed to create sentence segments suitable
-for speech recognition training, and to filter out segments with low-quality transcription. For system training,
-GigaSpeech provides five subsets of different sizes, 10h, 250h, 1000h, 2500h, and 10000h.
-For our 10,000-hour XL training subset, we cap the word error rate at 4% during the filtering/validation stage,
-and for all our other smaller training subsets, we cap it at 0%. The DEV and TEST evaluation sets, on the other hand,
-are re-processed by professional human transcribers to ensure high transcription quality.
+The AMI Meeting Corpus consists of 100 hours of meeting recordings. The recordings use a range of signals
+synchronized to a common timeline. These include close-talking and far-field microphones, individual and
+room-view video cameras, and output from a slide projector and an electronic whiteboard. During the meetings,
+the participants also have unsynchronized pens available to them that record what is written. The meetings
+were recorded in English using three different rooms with different acoustic properties, and include mostly
+non-native speakers. \n
 """
 
 _HOMEPAGE = "https://groups.inf.ed.ac.uk/ami/corpus/"
@@ -263,6 +249,12 @@ _EVAL_SAMPLE_IDS = [
     "TS3003d",
 ]
 
+_SAMPLE_IDS = {
+    "train": _TRAIN_SAMPLE_IDS,
+    "dev": _VALIDATION_SAMPLE_IDS,
+    "eval": _EVAL_SAMPLE_IDS,
+}
+
 _SUBSETS = ("ihm",)
 
 _BASE_DATA_URL = "https://huggingface.co/datasets/edinburghcstr/ami/resolve/main/"
@@ -283,19 +275,6 @@ class AMIConfig(datasets.BuilderConfig):
 
 
 class AMI(datasets.GeneratorBasedBuilder):
-    """
-    GigaSpeech is an evolving, multi-domain English speech recognition corpus with 10,000 hours of high quality
-    labeled audio suitable for supervised training, and 40,000 hours of total audio suitable for semi-supervised
-    and unsupervised training (this implementation contains only labelled data for now).
-    Around 40,000 hours of transcribed audio is first collected from audiobooks, podcasts
-    and YouTube, covering both read and spontaneous speaking styles, and a variety of topics, such as arts, science,
-    sports, etc. A new forced alignment and segmentation pipeline is proposed to create sentence segments suitable
-    for speech recognition training, and to filter out segments with low-quality transcription. For system training,
-    GigaSpeech provides five subsets of different sizes, 10h, 250h, 1000h, 2500h, and 10000h.
-    For our 10,000-hour XL training subset, we cap the word error rate at 4% during the filtering/validation stage,
-    and for all our other smaller training subsets, we cap it at 0%. The DEV and TEST evaluation sets, on the other hand,
-    are re-processed by professional human transcribers to ensure high transcription quality.
-    """
 
     VERSION = datasets.Version("1.0.0")
 
@@ -308,7 +287,7 @@ class AMI(datasets.GeneratorBasedBuilder):
     def _info(self):
         features = datasets.Features(
             {
-                "segment_id": datasets.Value("string"),
+                "meeting_id": datasets.Value("string"),
                 "audio_id": datasets.Value("string"),
                 "text": datasets.Value("string"),
                 "audio": datasets.Audio(sampling_rate=16_000),
@@ -327,46 +306,68 @@ class AMI(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
-        train_audio_files = {m: _AUDIO_ARCHIVE_URL.format(subset=self.config.name, split="train", _id=m) for m in _TRAIN_SAMPLE_IDS}
-        dev_audio_files = {m: _AUDIO_ARCHIVE_URL.format(subset=self.config.name, split="dev", _id=m) for m in _VALIDATION_SAMPLE_IDS}
-        eval_audio_files = {m: _AUDIO_ARCHIVE_URL.format(subset=self.config.name, split="eval", _id=m) for m in _EVAL_SAMPLE_IDS}
+        splits = ["train", "dev", "eval"]
+
+        audio_archives_urls = {}
+        for split in splits:
+            audio_archives_urls[split] = [
+                _AUDIO_ARCHIVE_URL.format(subset=self.config.name, split=split, _id=m) for m in _SAMPLE_IDS[split]
+            ]
 
-        train_audio_archives = dl_manager.download_and_extract(train_audio_files)
-        dev_audio_archives = dl_manager.download_and_extract(dev_audio_files)
-        eval_audio_archives = dl_manager.download_and_extract(eval_audio_files)
+        audio_archives = dl_manager.download(audio_archives_urls)
+        local_extracted_archives_paths = dl_manager.extract(audio_archives) if not dl_manager.is_streaming else {
+            split: [None] * len(audio_archives[split]) for split in splits
+        }
 
-        train_annotation = dl_manager.download_and_extract(_ANNOTATIONS_ARCHIVE_URL.format(split="train"))
-        dev_annotation = dl_manager.download_and_extract(_ANNOTATIONS_ARCHIVE_URL.format(split="dev"))
-        eval_annotation = dl_manager.download_and_extract(_ANNOTATIONS_ARCHIVE_URL.format(split="eval"))
+        annotations_urls = {split: _ANNOTATIONS_ARCHIVE_URL.format(split=split) for split in splits}
+        annotations = dl_manager.download(annotations_urls)
 
         return [
             datasets.SplitGenerator(
                 name=datasets.Split.TRAIN,
-                gen_kwargs={"audio": train_audio_archives, "annotation": train_annotation, "split": "train"},
+                gen_kwargs={
+                    "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_archives["train"]],
+                    "local_extracted_archives_paths": local_extracted_archives_paths["train"],
+                    "annotation": annotations["train"],
+                    "split": "train"
+                },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.VALIDATION,
-                gen_kwargs={"audio": dev_audio_archives, "annotation": dev_annotation, "split": "dev"},
+                gen_kwargs={
+                    "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_archives["dev"]],
+                    "local_extracted_archives_paths": local_extracted_archives_paths["dev"],
+                    "annotation": annotations["dev"],
+                    "split": "dev"
+                },
             ),
             datasets.SplitGenerator(
                 name=datasets.Split.TEST,
-                gen_kwargs={"audio": eval_audio_archives, "annotation": eval_annotation, "split": "eval"},
+                gen_kwargs={
+                    "audio_archives": [dl_manager.iter_archive(archive) for archive in audio_archives["eval"]],
+                    "local_extracted_archives_paths": local_extracted_archives_paths["eval"],
+                    "annotation": annotations["eval"],
+                    "split": "eval"
+                },
             ),
         ]
 
-    def _generate_examples(self, audio, annotation, split):
+    def _generate_examples(self, audio_archives, local_extracted_archives_paths, annotation, split):
         # open annotation file
+        assert len(audio_archives) == len(local_extracted_archives_paths)
+
         with open(annotation, "r", encoding="utf-8") as f:
             transcriptions = {}
             for line in f.readlines():
                 line_items = line.strip().split()
                 _id = line_items[0]
                 text = " ".join(line_items[1:])
-                _, segment_id, microphone_id, speaker_id, begin_time, end_time = _id.split("_")
+                _, meeting_id, microphone_id, speaker_id, begin_time, end_time = _id.split("_")
+                audio_filename = "_".join([split, _id.lower()]) + ".wav"
 
-                transcriptions[_id] = {
+                transcriptions[audio_filename] = {
                     "audio_id": _id,
-                    "segment_id": segment_id,
+                    "meeting_id": meeting_id,
                     "text": text,
                     "begin_time": int(begin_time) / 100,
                     "end_time": int(end_time) / 100,
@@ -374,10 +375,16 @@ class AMI(datasets.GeneratorBasedBuilder):
                     "speaker_id": speaker_id,
                 }
 
-        for _audio_id, (transcription_id, result) in enumerate(transcriptions.items()):
-            folder_id = result["segment_id"]
-            file_name = "_".join([split, transcription_id.lower()]) + ".wav"
-            audio_file = os.path.join(audio[folder_id], folder_id, file_name)
-            result["audio"] = audio_file
-            yield _audio_id, result
+        features = ["meeting_id", "audio_id", "text", "begin_time", "end_time", "microphone_id", "speaker_id"]
+        for archive, local_archive_path in zip(audio_archives, local_extracted_archives_paths):
+            for audio_path, audio_file in archive:
+                # audio_path is like 'EN2001a/train_ami_en2001a_h00_mee068_0414915_0415078.wav'
+                audio_meta = transcriptions[audio_path.split("/")[-1]]
 
+                yield audio_path, {
+                    "audio": {
+                        "path": os.path.join(local_archive_path, audio_path) if local_archive_path else audio_path,
+                        "bytes": audio_file.read(),
+                    },
+                    **{feature: audio_meta[feature] for feature in features}
+                }