ami.py

# coding=utf-8
# Copyright 2021 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

# Lint as: python3
"""AMI Corpus"""

import os
import xml.etree.ElementTree as ET

import numpy as np

import datasets


logger = datasets.logging.get_logger(__name__)

_CITATION = """\
@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}
}
"""

_URL = "https://groups.inf.ed.ac.uk/ami/corpus/"

_DL_URL_ANNOTATIONS = "http://groups.inf.ed.ac.uk/ami/AMICorpusAnnotations/ami_public_manual_1.6.2.zip"
_DL_SAMPLE_FORMAT = "https://groups.inf.ed.ac.uk/ami/AMICorpusMirror//amicorpus/{}/audio/{}"

_SPEAKERS = ["A", "B", "C", "D", "E"]

# Commented out samples don't seem to exist

_TRAIN_SAMPLE_IDS = [
    "ES2002a",
    "ES2002b",
    "ES2002c",
    "ES2002d",
    "ES2003a",
    "ES2003b",
    "ES2003c",
    "ES2003d",
    "ES2005a",
    "ES2005b",
    "ES2005c",
    "ES2005d",
    "ES2006a",
    "ES2006b",
    "ES2006c",
    "ES2006d",
    "ES2007a",
    "ES2007b",
    "ES2007c",
    "ES2007d",
    "ES2008a",
    "ES2008b",
    "ES2008c",
    "ES2008d",
    "ES2009a",
    "ES2009b",
    "ES2009c",
    "ES2009d",
    "ES2010a",
    "ES2010b",
    "ES2010c",
    "ES2010d",
    "ES2012a",
    "ES2012b",
    "ES2012c",
    "ES2012d",
    "ES2013a",
    "ES2013b",
    "ES2013c",
    "ES2013d",
    "ES2014a",
    "ES2014b",
    "ES2014c",
    "ES2014d",
    "ES2015a",
    "ES2015b",
    "ES2015c",
    "ES2015d",
    "ES2016a",
    "ES2016b",
    "ES2016c",
    "ES2016d",
    "IS1000a",
    "IS1000b",
    "IS1000c",
    "IS1000d",
    "IS1001a",
    "IS1001b",
    "IS1001c",
    "IS1001d",
    "IS1002b",
    "IS1002c",
    "IS1002d",
    "IS1003a",
    "IS1003b",
    "IS1003c",
    "IS1003d",
    "IS1004a",
    "IS1004b",
    "IS1004c",
    "IS1004d",
    "IS1005a",
    "IS1005b",
    "IS1005c",
    "IS1006a",
    "IS1006b",
    "IS1006c",
    "IS1006d",
    "IS1007a",
    "IS1007b",
    "IS1007c",
    "IS1007d",
    "TS3005a",
    "TS3005b",
    "TS3005c",
    "TS3005d",
    "TS3006a",
    "TS3006b",
    "TS3006c",
    "TS3006d",
    "TS3007a",
    "TS3007b",
    "TS3007c",
    "TS3007d",
    "TS3008a",
    "TS3008b",
    "TS3008c",
    "TS3008d",
    "TS3009a",
    "TS3009b",
    "TS3009c",
    "TS3009d",
    "TS3010a",
    "TS3010b",
    "TS3010c",
    "TS3010d",
    "TS3011a",
    "TS3011b",
    "TS3011c",
    "TS3011d",
    "TS3012a",
    "TS3012b",
    "TS3012c",
    "TS3012d",
    "EN2001a",
    "EN2001b",
    "EN2001d",
    "EN2001e",
    "EN2003a",
    "EN2004a",
    "EN2005a",
    "EN2006a",
    "EN2006b",
    "EN2009b",
    "EN2009c",
    "EN2009d",
    "IN1001",
    "IN1002",
    "IN1005",
    "IN1007",
    "IN1008",
    "IN1009",
    "IN1012",
    "IN1013",
    "IN1014",
    "IN1016",
]

_VALIDATION_SAMPLE_IDS = [
    "ES2011a",
    "ES2011b",
    "ES2011c",
    "ES2011d",
    "IS1008a",
    "IS1008b",
    "IS1008c",
    "IS1008d",
    "TS3004a",
    "TS3004b",
    "TS3004c",
    "TS3004d",
    "IB4001",
    "IB4002",
    "IB4003",
    "IB4004",
    "IB4010",
    "IB4011",
]


_EVAL_SAMPLE_IDS = [
    "ES2004a",
    "ES2004b",
    "ES2004c",
    "ES2004d",
    "IS1009a",
    "IS1009b",
    "IS1009c",
    "IS1009d",
    "TS3003a",
    "TS3003b",
    "TS3003c",
    "TS3003d",
    "EN2002a",
    "EN2002b",
    "EN2002c",
    "EN2002d",
]


_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. \n
"""


class AMIConfig(datasets.BuilderConfig):
    """BuilderConfig for LibriSpeechASR."""

    def __init__(self, formats, missing_files=None, **kwargs):
        """
        Args:
          formats: `List[string]`, a list of audio file formats
          missing_files: `List[string]`, a list of missing audio file ids
          **kwargs: keyword arguments forwarded to super.
        """
        self.dl_path_formats = [_DL_SAMPLE_FORMAT + "." + f + ".wav" for f in formats]

        # for microphone configs some audio files are missing
        self.missing_files = missing_files if missing_files is not None else []
        super(AMIConfig, self).__init__(version=datasets.Version("1.6.2", ""), **kwargs)


class AMI(datasets.GeneratorBasedBuilder):
    """AMI dataset."""

    BUILDER_CONFIGS = [
        AMIConfig(name="headset-single", formats=["Mix-Headset"], description=""),
        AMIConfig(name="headset-multi", formats=["Headset-0", "Headset-1", "Headset-2", "Headset-3"], description=""),
        AMIConfig(
            name="microphone-single",
            formats=["Array1-01"],
            missing_files=["IS1003b", "IS1007d"],
        ),
        AMIConfig(
            name="microphone-multi",
            formats=[
                "Array1-01",
                "Array1-02",
                "Array1-03",
                "Array1-04",
                "Array1-05",
                "Array1-06",
                "Array1-07",
                "Array1-08",
            ],
            missing_files=["IS1003b", "IS1007d"],
        ),
    ]

    def _info(self):
        features_dict = {
            "word_ids": datasets.Sequence(datasets.Value("string")),
            "word_start_times": datasets.Sequence(datasets.Value("float")),
            "word_end_times": datasets.Sequence(datasets.Value("float")),
            "word_speakers": datasets.Sequence(datasets.Value("string")),
            "segment_ids": datasets.Sequence(datasets.Value("string")),
            "segment_start_times": datasets.Sequence(datasets.Value("float")),
            "segment_end_times": datasets.Sequence(datasets.Value("float")),
            "segment_speakers": datasets.Sequence(datasets.Value("string")),
            "words": datasets.Sequence(datasets.Value("string")),
            "channels": datasets.Sequence(datasets.Value("string")),
        }

        if self.config.name == "headset-single":
            features_dict.update({"file": datasets.Value("string")})
            features_dict.update({"audio": datasets.features.Audio(sampling_rate=16_000)})
            config_description = (
                "Close talking audio of single headset. "
                "This configuration only includes audio belonging to the "
                "headset of the person currently speaking."
            )
        elif self.config.name == "microphone-single":
            features_dict.update({"file": datasets.Value("string")})
            features_dict.update({"audio": datasets.features.Audio(sampling_rate=16_000)})
            config_description = (
                "Far field audio of single microphone. "
                "This configuration only includes audio belonging the first microphone, "
                "*i.e.* 1-1, of the microphone array."
            )
        elif self.config.name == "headset-multi":
            features_dict.update({f"file-{i}": datasets.Value("string") for i in range(4)})
            features_dict.update({f"file-{i}": datasets.features.Audio(sampling_rate=16_000) for i in range(4)})
            config_description = (
                "Close talking audio of four individual headset. "
                "This configuration includes audio belonging to four individual headsets."
                " For each annotation there are 4 audio files 0, 1, 2, 3."
            )
        elif self.config.name == "microphone-multi":
            features_dict.update({f"file-1-{i}": datasets.Value("string") for i in range(1, 8)})
            features_dict.update({f"file-1-{i}": datasets.features.Audio(sampling_rate=16_000) for i in range(1, 8)})
            config_description = (
                "Far field audio of microphone array. "
                "This configuration includes audio of "
                "the first microphone array 1-1, 1-2, ..., 1-8."
            )
        else:
            raise ValueError(f"Configuration {self.config.name} does not exist.")

        return datasets.DatasetInfo(
            description=_DESCRIPTION + config_description,
            features=datasets.Features(features_dict),
            homepage=_URL,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):

        # multi-processing doesn't work for AMI
        if hasattr(dl_manager, "_download_config") and dl_manager._download_config.num_proc != 1:
            logger.warning(
                "AMI corpus cannot be downloaded using multi-processing. "
                "Setting number of downloaded processes `num_proc` to 1. "
            )
            dl_manager._download_config.num_proc = 1

        annotation_path = dl_manager.download_and_extract(_DL_URL_ANNOTATIONS)

        # train
        train_files = [path.format(_id, _id) for _id in _TRAIN_SAMPLE_IDS for path in self.config.dl_path_formats]
        train_files = list(
            filter(lambda f: f.split("/")[-1].split(".")[0] not in self.config.missing_files, train_files)
        )
        train_ids = [f.split("/")[-1].split(".")[0] for f in train_files]
        train_path = dl_manager.download_and_extract(train_files)

        # validation
        validation_files = [
            path.format(_id, _id) for _id in _VALIDATION_SAMPLE_IDS for path in self.config.dl_path_formats
        ]
        validation_ids = [f.split("/")[-1].split(".")[0] for f in validation_files]
        validation_path = dl_manager.download_and_extract(validation_files)

        # test
        eval_files = [path.format(_id, _id) for _id in _EVAL_SAMPLE_IDS for path in self.config.dl_path_formats]
        eval_ids = [f.split("/")[-1].split(".")[0] for f in eval_files]
        eval_path = dl_manager.download_and_extract(eval_files)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "annotation_path": annotation_path,
                    "samples_paths": train_path,
                    "ids": _TRAIN_SAMPLE_IDS,
                    "verification_ids": train_ids,
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "annotation_path": annotation_path,
                    "samples_paths": validation_path,
                    "ids": _VALIDATION_SAMPLE_IDS,
                    "verification_ids": validation_ids,
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "annotation_path": annotation_path,
                    "samples_paths": eval_path,
                    "ids": _EVAL_SAMPLE_IDS,
                    "verification_ids": eval_ids,
                },
            ),
        ]

    @staticmethod
    def _sort(key, lists):
        indices = np.argsort(key)

        sorted_lists = [np.array(array)[indices].tolist() for array in lists]
        return sorted_lists

    @staticmethod
    def _extract_words_annotations(paths):
        word_ids = []
        word_start_times = []
        word_end_times = []
        words = []
        word_speakers = []

        for path in paths:
            # retrive speaker
            speaker = path.split(".")[-3]

            with open(path, "r", encoding="utf-8") as words_file:
                root = ET.parse(words_file).getroot()
                for type_tag in root.findall("w"):
                    word_id = type_tag.get("{http://nite.sourceforge.net/}id")

                    word_start_time = type_tag.get("starttime")
                    word_end_time = type_tag.get("endtime")
                    text = type_tag.text

                    if word_start_time is not None and word_end_time is not None:
                        word_ids.append(word_id)
                        word_start_times.append(float(word_start_time))
                        word_end_times.append(float(word_end_time))
                        words.append(text)
                        word_speakers.append(speaker)
                    else:
                        logger.warning(
                            f"Annotation {word_id} of file {path} is missing information about"
                            "either word_start_time or word_end_time. Skipping sample..."
                        )

        return AMI._sort(word_start_times, [word_ids, word_start_times, word_end_times, words, word_speakers])

    @staticmethod
    def _extract_segments_annotations(paths):
        segment_ids = []
        channels = []
        segment_start_times = []
        segment_end_times = []
        segment_speakers = []

        for path in paths:
            speaker = path.split(".")[-3]

            with open(path, "r", encoding="utf-8") as segments_file:
                root = ET.parse(segments_file).getroot()
                for type_tag in root.findall("segment"):
                    segment_ids.append(type_tag.get("{http://nite.sourceforge.net/}id"))
                    segment_start_times.append(float(type_tag.get("transcriber_start")))
                    segment_end_times.append(float(type_tag.get("transcriber_end")))
                    channels.append(type_tag.get("channel"))
                    segment_speakers.append(speaker)

        return AMI._sort(
            segment_start_times, [segment_ids, segment_start_times, segment_end_times, channels, segment_speakers]
        )

    def _generate_examples(self, annotation_path, samples_paths, ids, verification_ids):
        logger.info(f"⏳ Generating {', '.join(ids)}")

        # number of audio files of config
        num_audios = len(self.config.dl_path_formats)

        # filter missing ids
        ids = list(filter(lambda n: n not in self.config.missing_files, ids))

        # audio
        samples_paths_dict = {}
        for i, _id in enumerate(ids):
            sample_paths = samples_paths[num_audios * i : num_audios * (i + 1)]
            sample_verification_id = set(verification_ids[num_audios * i : num_audios * (i + 1)])

            # make sure that multi microphone samples are correctly atttributed to labels
            if len(sample_verification_id) > 1 or next(iter(sample_verification_id)) != _id:
                raise ValueError(
                    f"Incorrect dataset generation. The files {sample_paths} of id {_id} have incorrect verification_ids {sample_verification_id}."
                )

            # set correct files correctly
            samples_paths_dict[_id] = sample_paths

        # words
        words_paths = {
            _id: [os.path.join(annotation_path, "words/{}.{}.words.xml".format(_id, speaker)) for speaker in _SPEAKERS]
            for _id in ids
        }
        words_paths = {_id: list(filter(lambda path: os.path.isfile(path), words_paths[_id])) for _id in ids}
        words_paths = {key: words_paths[key] for key in words_paths if len(words_paths[key]) > 0}

        # segments
        segments_paths = {
            _id: [
                os.path.join(annotation_path, "segments/{}.{}.segments.xml".format(_id, speaker))
                for speaker in _SPEAKERS
            ]
            for _id in ids
        }
        segments_paths = {_id: list(filter(lambda path: os.path.isfile(path), segments_paths[_id])) for _id in ids}
        segments_paths = {key: segments_paths[key] for key in segments_paths if len(segments_paths[key]) > 0}

        for _id in words_paths.keys():
            word_ids, word_start_times, word_end_times, words, word_speakers = self._extract_words_annotations(
                words_paths[_id]
            )

            (
                segment_ids,
                segment_start_times,
                segment_end_times,
                channels,
                segment_speakers,
            ) = self._extract_segments_annotations(segments_paths[_id])

            result = {
                "word_ids": word_ids,
                "word_start_times": word_start_times,
                "word_end_times": word_end_times,
                "word_speakers": word_speakers,
                "segment_ids": segment_ids,
                "segment_start_times": segment_start_times,
                "segment_end_times": segment_end_times,
                "segment_speakers": segment_speakers,
                "channels": channels,
                "words": words,
            }

            if self.config.name in ["headset-single", "microphone-single"]:
                result.update({"file": samples_paths_dict[_id][0], "audio": samples_paths_dict[_id][0]})
            elif self.config.name in ["headset-multi"]:
                result.update({f"file-{i}": samples_paths_dict[_id][i] for i in range(num_audios)})
                result.update({f"audio-{i}": samples_paths_dict[_id][i] for i in range(num_audios)})
            elif self.config.name in ["microphone-multi"]:
                result.update({f"file-1-{i+1}": samples_paths_dict[_id][i] for i in range(num_audios)})
                result.update({f"audio-1-{i+1}": samples_paths_dict[_id][i] for i in range(num_audios)})
            else:
                raise ValueError(f"Configuration {self.config.name} does not exist.")

            yield _id, result