Datasets:

ArXiv:
License:
File size: 8,202 Bytes
b036932
 
 
 
 
 
 
 
 
 
 
 
 
 
00846d9
 
 
 
 
 
b036932
 
2201c7f
b036932
 
2201c7f
b036932
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a44f1ba
 
00846d9
 
 
 
 
b036932
00846d9
 
 
b036932
a44f1ba
 
2201c7f
a44f1ba
 
 
2201c7f
 
 
 
0e959bb
 
 
 
b036932
 
00846d9
b036932
 
 
 
 
 
a44f1ba
b036932
a44f1ba
 
b036932
 
 
 
00846d9
 
b036932
 
 
 
a44f1ba
2201c7f
 
b036932
 
 
 
 
 
2201c7f
 
 
 
 
 
 
b036932
 
 
 
00846d9
 
 
b036932
 
00846d9
b036932
 
 
a44f1ba
2201c7f
 
 
a44f1ba
2201c7f
b036932
0e959bb
2201c7f
 
a44f1ba
 
0e959bb
 
b036932
 
937819b
b036932
a44f1ba
2201c7f
 
0e959bb
b036932
 
 
 
0e959bb
 
 
 
 
 
 
2201c7f
 
 
 
 
 
 
 
 
 
 
 
a44f1ba
b036932
2201c7f
2ed7507
b036932
2201c7f
b036932
 
 
a44f1ba
 
b036932
 
 
00846d9
0e959bb
 
 
d1ba203
0e959bb
2201c7f
 
 
 
 
 
 
 
 
 
 
a44f1ba
 
 
2201c7f
b036932
 
 
2201c7f
b036932
 
 
2201c7f
b036932
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
# coding=utf-8
# Copyright 2022 The PolyAI and 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.
# 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.
"""
EVI is a challenging spoken multilingual dataset with 5,506 dialogues in English, Polish, and French
that can be used for benchmarking and developing knowledge-based enrolment, identification, and identification
for spoken dialogue systems.
"""


import csv
from datetime import datetime
import json
import os
import warnings

import datasets

logger = datasets.logging.get_logger(__name__)


_CITATION = """\
@inproceedings{Spithourakis2022evi,
    author      = {Georgios P. Spithourakis and Ivan Vuli\'{c} and Micha\l{} Lis and I\~{n}igo Casanueva and Pawe\l{} Budzianowski},
    title       = {{EVI}: Multilingual Spoken Dialogue Tasks and Dataset for Knowledge-Based Enrolment, Verification, and Identification},
    year        = {2022},
    note        = {Data available at https://github.com/PolyAI-LDN/evi-paper},
    url         = {https://arxiv.org/abs/2204.13496},
    booktitle   = {Findings of NAACL (publication pending)}
}
"""  # noqa

_ALL_CONFIGS = sorted([
    "en-GB", "fr-FR", "pl-PL"
])

_LANGS = sorted(["en", "fr", "pl"])

_DESCRIPTION = """
EVI is a challenging spoken multilingual dataset with 5,506 dialogues in English, Polish, and French 
that can be used for benchmarking and developing knowledge-based enrolment, identification, and identification 
for spoken dialogue systems.
"""  # noqa

_LICENSE = "CC-BY-4.0"

_HOMEPAGE = "https://github.com/PolyAI-LDN/evi-paper"

_BASE_URL = "https://huggingface.co/datasets/PolyAI/evi/resolve/main/data"

_TEXTS_URL = {
    lang: os.path.join(_BASE_URL, f"dialogues.{lang.split('-')[0]}.tsv") for lang in _LANGS
}

_RECORDS_URL = {
    lang: os.path.join(_BASE_URL, f"records.{lang.split('-')[0]}.csv") for lang in _LANGS
}

_BROKEN_URL = {
    "en": os.path.join(_BASE_URL, "broken_en.txt")
}

_AUDIO_DATA_URL = "https://poly-public-data.s3.eu-west-2.amazonaws.com/evi-paper/audios.zip"  # noqa

_VERSION = datasets.Version("0.0.1", "")


class EviConfig(datasets.BuilderConfig):
    """BuilderConfig for EVI"""

    def __init__(
        self, name, *args, **kwargs
    ):
        super().__init__(name=name, *args, **kwargs)
        self.languages = _LANGS if name == "all" else [name.split("-")[0]]  # all langs if config == "all"


class Evi(datasets.GeneratorBasedBuilder):

    DEFAULT_WRITER_BATCH_SIZE = 512
    BUILDER_CONFIGS = [EviConfig(name) for name in _ALL_CONFIGS + ["all"]]

    def _info(self):
        features = datasets.Features(
            {
                "language": datasets.ClassLabel(names=_LANGS),
                "audio": datasets.Audio(sampling_rate=8_000),
                "asr_transcription": datasets.Value("string"),
                "dialogue_id": datasets.Value("string"),
                "speaker_id": datasets.Value("string"),
                "turn_id": datasets.Value("int32"),
                "target_profile_id": datasets.Value("string"),
                "asr_nbest": datasets.Sequence(datasets.Value("string")),
                "path": datasets.Value("string"),
                "postcode": datasets.Value("string"),
                "name": datasets.Value("string"),
                "dob": datasets.Value("date64"),
                "name_first": datasets.Value("string"),
                "name_last": datasets.Value("string"),
                "sex": datasets.ClassLabel(names=["F", "M"]),  # TODO: are there other genders or Nones?
                "email": datasets.Value("string"),
            }
        )

        return datasets.DatasetInfo(
            version=_VERSION,
            description=_DESCRIPTION,
            license=_LICENSE,
            citation=_CITATION,
            features=features,
            homepage=_HOMEPAGE
        )

    def _split_generators(self, dl_manager):
        langs = self.config.languages
        lang2records_urls = {
            lang: _RECORDS_URL[lang] for lang in langs
        }
        lang2text_urls = {
            lang: _TEXTS_URL[lang] for lang in langs
        }

        records_paths = dl_manager.download_and_extract(lang2records_urls)
        text_paths = dl_manager.download_and_extract(lang2text_urls)
        audio_data_path = dl_manager.download_and_extract(_AUDIO_DATA_URL)

        broken_path = dl_manager.download_and_extract(_BROKEN_URL["en"]) if "en" in langs else None

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "audio_data_path": audio_data_path,
                    "text_paths": text_paths,
                    "records_paths": records_paths,
                    "broken_path": broken_path
                },
            )
        ]

    def _generate_examples(self, audio_data_path, text_paths, records_paths, broken_path=None):
        if broken_path:
            with open(broken_path, encoding="utf-8") as f:
                broken_samples = set([line.strip() for line in f])
        else:
            broken_samples = None

        for lang, text_path in text_paths.items():

            records_path = records_paths[lang]
            records = dict()
            with open(records_path, encoding="utf-8") as fin:
                records_reader = csv.DictReader(
                    fin, delimiter=",", skipinitialspace=True
                )
                for row in records_reader:
                    records[row["scenario_id"]] = row
                    records[row["scenario_id"]]["dob"] = datetime.strptime(row["dob"], "%Y-%m-%d")
                    _ = records[row["scenario_id"]].pop("scenario_id")

            with open(text_path, encoding="utf-8") as fin:
                texts_reader = csv.DictReader(
                    fin, delimiter="\t", skipinitialspace=True
                )
                for dictrow in texts_reader:
                    dialogue_id = dictrow["dialogue_id"]
                    turn_id = dictrow["turn_num"]
                    file_path = os.path.join(
                        "audios",
                        lang,
                        dialogue_id,
                        f'{turn_id}.wav'
                    )
                    full_path = os.path.join(audio_data_path, file_path)
                    if broken_samples and file_path in broken_samples:
                        warnings.warn(f"{full_path} is broken, skipping it.")
                        continue
                    if not os.path.isfile(full_path):
                        warnings.warn(f"{full_path} not found, skipping it.")
                        continue

                    target_profile_id = dictrow["scenario_id"]
                    if target_profile_id not in records:
                        warnings.warn(
                            f"""
                            Record with scenario_id {target_profile_id} not found, ignoring this dialogue. 
                            Full dialogue info: {dictrow}
                            """
                        )
                        continue

                    yield file_path, {
                        "language": lang,
                        "audio": str(full_path),
                        "dialogue_id": dialogue_id,
                        "speaker_id": dictrow["speaker_id"],
                        "turn_id": turn_id,
                        "target_profile_id": target_profile_id,
                        "asr_transcription": dictrow["transcription"],
                        "asr_nbest": json.loads(dictrow["nbest"]),
                        "path": file_path,
                        **records[target_profile_id]
                    }