Duplicate from Shiry/ATC_combined

Co-authored-by: Yonash <Shiry@users.noreply.huggingface.co>

.gitattributes

@@ -0,0 +1,54 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,143 @@
+---
+dataset_info:
+  features:
+  - name: id
+    dtype: string
+  - name: audio
+    dtype:
+      audio:
+        sampling_rate: 16000
+  - name: text
+    dtype: string
+  - name: segment_start_time
+    dtype: float32
+  - name: segment_end_time
+    dtype: float32
+  - name: duration
+    dtype: float32
+  splits:
+  - name: test
+    num_bytes: 612270626
+    num_examples: 4723
+  - name: train
+    num_bytes: 2543440112
+    num_examples: 18929
+tags:
+- audio
+- automatic-speech-recognition
+- en-atc
+- en
+- noisy-speech-recognition
+- speech-recognition
+task_categories:
+- automatic-speech-recognition  
+language:
+- en
+multilinguality:
+- monolingual
+license:
+- cc-by-nc-sa-4.0
+---
+
+# Dataset Card for UWB-ATCC corpus
+
+## Table of Contents
+- [Dataset Description](#dataset-description)
+  - [Dataset Summary](#dataset-summary)
+  - [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
+  - [Languages and Other Details](#languages-and-other-details)
+- [Dataset Structure](#dataset-structure)
+  - [Data Fields](#data-fields)
+- [Additional Information](#additional-information)
+  - [Licensing Information](#licensing-information)
+  - [Citation Information](#citation-information)
+
+
+## Dataset Description 
+- **Homepage:** [UWB-ATCC corpus homepage](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0)
+- **Repository:** [GitHub repository (used in research)](https://github.com/idiap/w2v2-air-traffic)
+- **Paper:** [Air traffic control communication (ATCC) speech corpora and their use for ASR and TTS development](https://link.springer.com/article/10.1007/s10579-019-09449-5)
+- **Paper of this research:** [How Does Pre-trained Wav2Vec 2.0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications](https://arxiv.org/abs/2203.16822)
+
+### Dataset Summary
+
+The UWB-ATCC Corpus is provided provided by University of West Bohemia, Department of Cybernetics. The corpus contains recordings of communication between air traffic controllers and pilots. The speech is manually transcribed and labeled with the information about the speaker (pilot/controller, not the full identity of the person). The corpus is currently small (20 hours) but we plan to search for additional data next year. The audio data format is: 8kHz, 16bit PCM, mono.
+
+Important, from the `<id (string)>` field, you can obtain the speaker roles. For instance:
+- `_PI`: segment with only pilot speech
+- `_AT`: segment with only ATCO speech
+- `PIAT`: segment with both, ATCO and pilot speech
+
+### Supported Tasks and Leaderboards
+
+- `automatic-speech-recognition`. Already adapted/fine-tuned models are available here --> [XLS-R-300m](https://huggingface.co/Jzuluaga/wav2vec2-large-960h-lv60-self-en-atc-atcosim).
+
+### Languages and other details
+
+The text and the recordings are in English. The authors took advantage of the fact that one of their industrial partners develops complex IT solutions for several ATC authorities and airports and, as such, has access to the ATC communication recordings collected in the Czech airspace. This partner was able to secure the following data:
+
+- Ground control—communication before takeoff and after landing—19.2 h of data.
+- Tower control—communication during takeoff, landing and landing standby—22.5 h.
+- Approach control—communication during landing approach—25.5 h.
+- Area control—communication during overflights and cruises—71.3 h.
+
+(Not all data is released. Check their website [here](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0))
+## Dataset Structure
+
+### Data Fields
+
+- `id (string)`: a string of recording identifier for each example, corresponding to its.
+- `audio (audio)`: audio data for the given ID
+- `text (string)`: transcript of the file already normalized. Follow these repositories for more details [w2v2-air-traffic](https://github.com/idiap/w2v2-air-traffic) and [bert-text-diarization-atc](https://github.com/idiap/bert-text-diarization-atc)
+- `segment_start_time (float32)`: segment start time (normally 0)
+- `segment_end_time (float32): segment end time
+- `duration (float32)`: duration of the recording, compute as segment_end_time - segment_start_time
+
+## Additional Information
+
+### Licensing Information
+
+The licensing status of the dataset hinges on the legal status of the [UWB-ATCC corpus](https://lindat.mff.cuni.cz/repository/xmlui/handle/11858/00-097C-0000-0001-CCA1-0) creators.
+
+They used [Creative Commons - Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0)](https://creativecommons.org/licenses/by-nc-sa/4.0/) licensing.
+
+### Citation Information
+
+Contributors who prepared, processed, normalized and uploaded the dataset in HuggingFace:
+
+```
+@article{zuluaga2022how,
+    title={How Does Pre-trained Wav2Vec2. 0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications},
+    author={Zuluaga-Gomez, Juan and Prasad, Amrutha and Nigmatulina, Iuliia and Sarfjoo, Saeed and others},
+    journal={IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar},
+    year={2022}
+  }
+
+@article{zuluaga2022bertraffic,
+  title={BERTraffic: BERT-based Joint Speaker Role and Speaker Change Detection for Air Traffic Control Communications},
+  author={Zuluaga-Gomez, Juan and Sarfjoo, Seyyed Saeed and Prasad, Amrutha and others},
+  journal={IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar},
+  year={2022}
+  }
+
+@article{zuluaga2022atco2,
+  title={ATCO2 corpus: A Large-Scale Dataset for Research on Automatic Speech Recognition and Natural Language Understanding of Air Traffic Control Communications},
+  author={Zuluaga-Gomez, Juan and Vesel{\`y}, Karel and Sz{\"o}ke, Igor and Motlicek, Petr and others},
+  journal={arXiv preprint arXiv:2211.04054},
+  year={2022}
+}
+```
+
+Authors of the dataset:
+```
+@article{vsmidl2019air,
+  title={Air traffic control communication (ATCC) speech corpora and their use for ASR and TTS development},
+  author={{\v{S}}m{\'\i}dl, Lubo{\v{s}} and {\v{S}}vec, Jan and Tihelka, Daniel and Matou{\v{s}}ek, Jind{\v{r}}ich and Romportl, Jan and Ircing, Pavel},
+  journal={Language Resources and Evaluation},
+  volume={53},
+  number={3},
+  pages={449--464},
+  year={2019},
+  publisher={Springer}
+}
+```

atc_data_loader.py

@@ -0,0 +1,275 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+#
+# SPDX-FileCopyrightText: Copyright © <2022> Idiap Research Institute <contact@idiap.ch>
+#
+# SPDX-FileContributor: Juan Zuluaga-Gomez <jzuluaga@idiap.ch>
+#
+# SPDX-License-Identifier: MIT-License
+
+"""\
+Script for loading air traffic control (ATC) speech datasets for automatic speech recognition (ASR).
+This script has been designed for ATC datasets that are in Kaldi format
+
+Required files: text, wav.scp and segments files
+
+- Databases
+    - Training: 
+        - ATCOSIM, LDC-ATCC and, UWB-ATCC corpora.
+    - Testing:
+        - ATCO2-test-set, ATCOSIM, LDC-ATCC and, UWB-ATCC corpora.
+"""
+
+import os
+import re
+
+import datasets
+import numpy as np
+import soundfile as sf
+from datasets.tasks import AutomaticSpeechRecognition
+
+_CITATION = """\
+@article{zuluaga2022does,
+  title={How Does Pre-trained Wav2Vec 2.0 Perform on Domain Shifted ASR? An Extensive Benchmark on Air Traffic Control Communications},
+  author={Zuluaga-Gomez, Juan and Prasad, Amrutha and Nigmatulina, Iuliia and Sarfjoo, Saeed and Motlicek, Petr and Kleinert, Matthias and Helmke, Hartmut and Ohneiser, Oliver and Zhan, Qingran},
+  journal={2022 IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar},
+  year={2022}
+}
+@article{zuluagabertraffic,
+  title={BERTraffic: BERT-based Joint Speaker Role and Speaker Change Detection for Air Traffic Control Communications (submitted to @ SLT-2022)},
+  author={Zuluaga-Gomez, Juan and Sarfjoo, Seyyed Saeed and Prasad, Amrutha and Nigmatulina, Iuliia and Motlicek, Petr and Ohneiser, Oliver and Helmke, Hartmut},
+  journal={2022 IEEE Spoken Language Technology Workshop (SLT), Doha, Qatar},
+  year={2022}
+}
+"""
+
+_DESCRIPTION = """\
+ATC speech DATASET. This DataLoader works with data in Kaldi format.
+    - We use the following files: text, segments and wav.scp 
+            - text --> utt_id transcript
+            - segments --> utt_id recording_id t_begin t_end
+            - wav.scp --> recording_id /path/to/wav/
+The default dataset is from ATCO2 project, a 1-hour sample: https://www.replaywell.com/atco2/download/ATCO2-ASRdataset-v1_beta.tgz
+"""
+
+_DATA_URL = "http://catalog.elra.info/en-us/repository/browse/ELRA-S0484/"
+
+_HOMEPAGE = "https://github.com/idiap/w2v2-air-traffic"
+
+logger = datasets.logging.get_logger(__name__)
+
+# Our models work with audio data at 16kHZ,
+_SAMPLING_RATE = int(16000)
+
+
+class ATCDataASRConfig(datasets.BuilderConfig):
+    """BuilderConfig for air traffic control datasets."""
+
+    def __init__(self, **kwargs):
+        """
+        Args:
+          data_dir: `string`, the path to the folder containing the files required to read: json or wav.scp
+          **kwargs: keyword arguments forwarded to super.
+        """
+        super(ATCDataASRConfig, self).__init__(**kwargs)
+
+
+class ATCDataASR(datasets.GeneratorBasedBuilder):
+
+    DEFAULT_WRITER_BATCH_SIZE = 256
+    DEFAULT_CONFIG_NAME = "all"
+    BUILDER_CONFIGS = [
+        # TRAIN, DEV AND TEST DATASETS
+        ATCDataASRConfig(name="train", description="ATC train dataset."),
+        ATCDataASRConfig(name="dev", description="ATC dev dataset."),
+        ATCDataASRConfig(name="test", description="ATC test dataset."),
+        # UNSUPERVISED DATASETS
+        ATCDataASRConfig(name="unsupervised", description="ATC unsupervised dataset."),
+    ]
+
+    # provide some information about the Dataset we just gathered
+    def _info(self):
+        return datasets.DatasetInfo(
+            description=_DESCRIPTION,
+            features=datasets.Features(
+                {
+                    "id": datasets.Value("string"),
+                    "file": datasets.Value("string"),
+                    "audio": datasets.features.Audio(sampling_rate=_SAMPLING_RATE),
+                    "text": datasets.Value("string"),
+                    "segment_start_time": datasets.Value("float"),
+                    "segment_end_time": datasets.Value("float"),
+                    "duration": datasets.Value("float"),
+                }
+            ),
+            supervised_keys=("audio", "text"),
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+            task_templates=[
+                AutomaticSpeechRecognition(
+                    audio_column="audio", transcription_column="text"
+                )
+            ],
+        )
+
+    def _split_generators(self, dlmanager):
+        """Returns SplitGenerators."""
+
+        split = self.config.name
+
+        # UNSUPERVISED set (used only for decoding)
+        if "unsupervised" in split:
+            split_name = datasets.Split.TEST
+        elif "test" in split or "dev" in split or "dummy" in split:
+            split_name = datasets.Split.TEST
+        # The last option left is: Train set
+        else:
+            split_name = datasets.Split.TRAIN
+
+        # you need to pass a data directory where the Kaldi folder is stored
+        filepath = self.config.data_dir
+
+        return [
+            datasets.SplitGenerator(
+                name=split_name,
+                # These kwargs will be passed to _generate_examples
+                gen_kwargs={
+                    "filepath": filepath,
+                    "split": split,
+                },
+            )
+        ]
+
+    def _generate_examples(self, filepath, split):
+        """You need to pass a path with the kaldi data, the folder should have
+        audio: wav.scp,
+        transcripts: text,
+        timing information: segments
+        """
+
+        logger.info("Generating examples located in: %s", filepath)
+
+        text_file = os.path.join(filepath, "text")
+        wavscp = os.path.join(filepath, "wav.scp")
+        segments = os.path.join(filepath, "segments")
+
+        id_ = ""
+        text_dict, wav_dict = {}, {}
+        segments_dict, utt2wav_id = {}, {}
+
+        line = 0
+        # get the text file
+        with open(text_file) as text_f:
+            for line in text_f:
+                if len(line.split(" ")) > 1:
+                    id_, transcript = line.split(" ", maxsplit=1)
+                    transcript = _remove_special_characters(transcript)
+                    if len(transcript.split(" ")) == 0:
+                        continue
+                    if len(transcript) < 2:
+                        continue
+                    text_dict[id_] = transcript
+                else:  # line is empty
+                    # if unsupervised set, then it's normal. else, continue
+                    if not "test_unsup" in self.config.name:
+                        continue
+                    id_ = line.rstrip().split(" ")[0]
+                    text_dict[id_] = ""
+
+        # get wav.scp and load data into memory
+        with open(wavscp) as text_f:
+            for line in text_f:
+                if line:
+                    if len(line.split()) < 2:
+                        continue
+                    id_, wavpath = line.split(" ", maxsplit=1)
+                    # only selects the part that ends of wav, flac or sph
+                    wavpath = [
+                        x
+                        for x in wavpath.split(" ")
+                        if ".wav" in x or ".WAV" in x or ".flac" in x or ".sph" in x
+                    ][0].rstrip()
+
+                    # make the output
+                    segment, sampling_rate = sf.read(wavpath, dtype=np.int16)
+                    wav_dict[id_] = [wavpath.rstrip(), segment, sampling_rate]
+
+        # get segments dictionary
+        with open(segments) as text_f:
+            for line in text_f:
+                if line:
+                    if len(line.split()) < 4:
+                        continue
+                    id_, wavid_, start, end = line.rstrip().split(" ")
+                    segments_dict[id_] = start.rstrip(), end.rstrip()
+                    utt2wav_id[id_] = wavid_
+
+        for rec_id, text in text_dict.items():
+            if rec_id in utt2wav_id and rec_id in segments_dict:
+
+                # get audio data from memory and the path of the file
+                wavpath, segment, sampling_rate = wav_dict[utt2wav_id[rec_id]]
+                # get timing information
+                seg_start, seg_end = segments_dict[rec_id]
+                seg_start, seg_end = float(seg_start), float(seg_end)
+                duration = round((seg_end - seg_start), 3)
+
+                # get the samples, bytes, already cropping by segment,
+                samples = _extract_audio_segment(
+                    segment, sampling_rate, float(seg_start), float(seg_end)
+                )
+
+                # output data for given dataset
+                example = {
+                    "audio": {
+                        "path": wavpath,
+                        "array": samples,
+                        "sampling_rate": sampling_rate,
+                    },
+                    "id": rec_id,
+                    "file": wavpath,
+                    "text": text,
+                    "segment_start_time": format(float(seg_start), ".3f"),
+                    "segment_end_time": format(float(seg_end), ".3f"),
+                    "duration": format(float(duration), ".3f"),
+                }
+
+                yield rec_id, example
+
+
+def _remove_special_characters(text):
+    """Function to remove some special chars/symbols from the given transcript"""
+
+    text = text.split(" ")
+    # first remove words between [] and <>
+    text = " ".join(
+        [
+            x
+            for x in text
+            if "[" not in x and "]" not in x and "<" not in x and ">" not in x
+        ]
+    )
+
+    # regex with predifined symbols to ignore/remove,
+    chars_to_ignore_regex2 = '[\{\[\]\<\>\/\,\?\.\!\u00AC\;\:"\\%\\\]|[0-9]'
+
+    text = re.sub(chars_to_ignore_regex2, "", text).lower()
+    sentence = text.replace("\u2013", "-")
+    sentence = sentence.replace("\u2014", "-")
+    sentence = sentence.replace("\u2018", "'")
+    sentence = sentence.replace("\u201C", "")
+    sentence = sentence.replace("\u201D", "")
+    sentence = sentence.replace("ñ", "n")
+    sentence = sentence.replace(" - ", " ")
+    sentence = sentence.replace("-", "")
+    sentence = sentence.replace("'", " ")
+    return sentence.lower().rstrip()
+
+
+def _extract_audio_segment(segment, sampling_rate, start_sec, end_sec):
+    """Extracts segment of audio samples (as an ndarray) from the given segment."""
+    # The dataset only contains mono audio.
+    start_sample = int(start_sec * sampling_rate)
+    end_sample = min(int(end_sec * sampling_rate), segment.shape[0])
+    samples = segment[start_sample:end_sample]
+    return samples

data/test-00000-of-00001-01544bdf54b4ccf3.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:52ad36731ede5f6cc528e95aa69e32c0f7be101646615590e86e7d84539c4653
+size 470060009

data/test-00000-of-00001-3a021115ca23c2a5.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a637e5b035c4e22bdf880554c900c3a775eac46c10dc9713e827dd6cbf38783
+size 131266472

data/train-00000-of-00002-4a9602acfde9f517.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0dcb6fbecc22116d74c8e079e367a379225b517874f5de803373389a9b0aa2b3
+size 302140417

data/train-00000-of-00004-c1d7fb31dcbf644a.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4948175e5ab8b3e221ad485854f73f47068891f61bba31f6360264186a45b08
+size 488301521

data/train-00001-of-00002-91082fb03180a296.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca9535c5ccf2d8a318b4835a9dfece6e00026d69b664e5002b16e8e83d0f5829
+size 278058025

data/train-00001-of-00004-f165730df6bf7253.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:855589e16589edbd45c21fb07dba5ff53ec7ceae1cbd187cbae413988e6c3a6b
+size 468054363

data/train-00002-of-00004-67e682f17e32b703.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9a3bb0a59d896f9b4897b5349936d0acd134a8965c9035ae766f50c8ce8b9d5e
+size 495020975

data/train-00003-of-00004-b0b05d4b243c95c6.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:333c1426084c3837d7b42c59b08618a9df1c25925dd636b8c0ad836dd641fea3
+size 473692438