Create croissant_rai_OpenWhistle-Pretraining.json

croissant_rai_OpenWhistle-Pretraining.json

@@ -0,0 +1,502 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "arrayShape": "cr:arrayShape",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "containedIn": "cr:containedIn",
+    "cr": "http://mlcommons.org/croissant/",
+    "data": {
+      "@id": "cr:data",
+      "@type": "@json"
+    },
+    "dataType": {
+      "@id": "cr:dataType",
+      "@type": "@vocab"
+    },
+    "dct": "http://purl.org/dc/terms/",
+    "extract": "cr:extract",
+    "equivalentProperty": "cr:equivalentProperty",
+    "examples": "cr:examples",
+    "field": "cr:field",
+    "fileObject": "cr:fileObject",
+    "fileProperty": "cr:fileProperty",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isArray": "cr:isArray",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "prov": "http://www.w3.org/ns/prov#",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "recordSet": "cr:recordSet",
+    "references": "cr:references",
+    "regex": "cr:regex",
+    "repeated": "cr:repeated",
+    "replace": "cr:replace",
+    "sc": "https://schema.org/",
+    "samplingRate": "cr:samplingRate",
+    "separator": "cr:separator",
+    "source": "cr:source",
+    "subField": "cr:subField",
+    "transform": "cr:transform"
+  },
+  "@type": "sc:Dataset",
+  "conformsTo": [
+    "http://mlcommons.org/croissant/1.1",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
+  "name": "OpenWhistle-Pretraining",
+  "alternateName": [
+    "OpenWhistleNeurIPS26/OpenWhistle-Pretraining",
+    "OpenWhistle-1.0-Pretraining"
+  ],
+  "description": "OpenWhistle 1.0 Pretraining is a public unlabeled 96 kHz dolphin acoustic audio dataset for self-supervised or unsupervised pretraining. It contains approximately 114 hours of raw audio corresponding to an estimated 180,000 dolphin whistles extracted from 33,267 sequences recorded over five years at Dolphin Reef, Eilat. The default configuration contains the full dataset, and the review-sample configuration provides a deterministic smaller subset for quick reviewer inspection.",
+  "url": "https://huggingface.co/datasets/OpenWhistleNeurIPS26/OpenWhistle-Pretraining",
+  "citeAs": "OpenWhistle: A Large-Scale Longitudinal Dataset and Benchmark of Bottlenose Dolphin Vocalizations. NeurIPS 2026 anonymous submission.",
+  "creator": {
+    "@type": "Organization",
+    "name": "Anonymous NeurIPS 2026 authors",
+    "url": "https://huggingface.co/OpenWhistleNeurIPS26"
+  },
+  "publisher": {
+    "@type": "Organization",
+    "name": "OpenWhistleNeurIPS26",
+    "url": "https://huggingface.co/OpenWhistleNeurIPS26"
+  },
+  "datePublished": "2026-04-17",
+  "dateModified": "2026-04-27",
+  "license": "https://creativecommons.org/licenses/by/4.0/",
+  "version": "1.0.0",
+  "identifier": "https://huggingface.co/datasets/OpenWhistleNeurIPS26/OpenWhistle-Pretraining",
+  "keywords": [
+    "dolphin",
+    "bioacoustics",
+    "audio",
+    "pretraining",
+    "self-supervised learning",
+    "96khz",
+    "hydrophone",
+    "Hugging Face",
+    "Croissant",
+    "Responsible AI"
+  ],
+  "isAccessibleForFree": true,
+  "isLiveDataset": false,
+  "spatialCoverage": {
+    "@type": "Place",
+    "name": "Dolphin Reef, Eilat, northern Gulf of Aqaba"
+  },
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "repo",
+      "name": "repo",
+      "description": "The Hugging Face git repository containing the dataset parquet conversion.",
+      "contentUrl": "https://huggingface.co/datasets/OpenWhistleNeurIPS26/OpenWhistle-Pretraining/tree/refs%2Fconvert%2Fparquet",
+      "encodingFormat": "git+https",
+      "sha256": "https://github.com/mlcommons/croissant/issues/80"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "parquet-files-for-config-default",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/x-parquet",
+      "includes": "default/*/*.parquet"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "parquet-files-for-config-review-sample",
+      "containedIn": {
+        "@id": "repo"
+      },
+      "encodingFormat": "application/x-parquet",
+      "includes": "review-sample/*/*.parquet"
+    }
+  ],
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "default_splits",
+      "name": "default_splits",
+      "description": "Splits for the default config.",
+      "dataType": "cr:Split",
+      "key": {
+        "@id": "default_splits/split_name"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "default_splits/split_name",
+          "dataType": "sc:Text"
+        }
+      ],
+      "data": [
+        {
+          "default_splits/split_name": "validation"
+        },
+        {
+          "default_splits/split_name": "train"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "default",
+      "name": "default",
+      "description": "Full OpenWhistle-Pretraining configuration. It has train and validation splits with 31,780 total examples, 114.284 hours of 96 kHz mono dolphin acoustic audio, and metadata columns for start_time, end_time, duration, year, and hydrophone.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "default/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": "default/(?:partial-)?(validation|train)/.+parquet$"
+            }
+          },
+          "references": {
+            "field": {
+              "@id": "default_splits/split_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/audio",
+          "dataType": "sc:AudioObject",
+          "samplingRate": 96000,
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "audio"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/start_time",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "start_time"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/end_time",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "end_time"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/duration",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "duration"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/year",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "year"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "default/hydrophone",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-default"
+            },
+            "extract": {
+              "column": "hydrophone"
+            }
+          }
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "review-sample_splits",
+      "name": "review-sample_splits",
+      "description": "Splits for the review-sample config.",
+      "dataType": "cr:Split",
+      "key": {
+        "@id": "review-sample_splits/split_name"
+      },
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample_splits/split_name",
+          "dataType": "sc:Text"
+        }
+      ],
+      "data": [
+        {
+          "review-sample_splits/split_name": "train"
+        },
+        {
+          "review-sample_splits/split_name": "validation"
+        }
+      ]
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "review-sample",
+      "name": "review-sample",
+      "description": "Deterministic review sample preserving the train and validation split names from the full dataset. It has 480 total examples and 1.850 hours of audio, intended for quick manual inspection without downloading the full dataset.",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/split",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "fileProperty": "fullpath"
+            },
+            "transform": {
+              "regex": "review\\-sample/(?:partial-)?(train|validation)/.+parquet$"
+            }
+          },
+          "references": {
+            "field": {
+              "@id": "review-sample_splits/split_name"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/audio",
+          "dataType": "sc:AudioObject",
+          "samplingRate": 96000,
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "audio"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/start_time",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "start_time"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/end_time",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "end_time"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/duration",
+          "dataType": "cr:Float32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "duration"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/year",
+          "dataType": "cr:Int32",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "year"
+            }
+          }
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "review-sample/hydrophone",
+          "dataType": "sc:Text",
+          "source": {
+            "fileSet": {
+              "@id": "parquet-files-for-config-review-sample"
+            },
+            "extract": {
+              "column": "hydrophone"
+            }
+          }
+        }
+      ]
+    }
+  ],
+  "rai:dataCollection": "The dataset was assembled from passive acoustic monitoring at Dolphin Reef, Eilat, a semi-natural marine site on the northern Gulf of Aqaba. Fixed hydrophones recorded a stable pod of known dolphins in a large natural marine area open to the sea. Human-dolphin interactions at the site are voluntary and initiated by the dolphins. Public examples are 96 kHz mono audio segments with timing metadata, recording year, and hydrophone channel. The public dataset covers recordings from 2019, 2020, 2021, 2023, and 2024, split into train and validation subsets.",
+  "rai:dataCollectionType": [
+    "Physical data collection",
+    "Direct measurement",
+    "Manual Human Curator",
+    "Software Collection"
+  ],
+  "rai:dataCollectionRawData": "Raw data consisted of underwater acoustic recordings captured by fixed and hidden hydrophones. The released data contains segmented audio examples and metadata fields: audio, start_time, end_time, duration, year, and hydrophone.",
+  "rai:dataCollectionMissingData": "The released pretraining dataset is unlabeled: it does not include whistle/noise labels, dolphin identity labels, species labels, behavioral context, environmental annotations, video context, exact hydrophone coordinates, or per-example animal identities. Coverage is uneven across recording years and hydrophone channels.",
+  "rai:dataPreprocessingProtocol": [
+    "Audio was segmented into whistle sequences and packaged as Hugging Face parquet datasets with an Audio feature at 96 kHz.",
+    "Rows shorter or unsuitable for pretraining may have been filtered during dataset preparation. Public metadata keeps segment start_time, end_time, duration, year, and hydrophone.",
+    "The review-sample configuration was generated as a deterministic shuffled subset using base seed 42, shuffle buffer size 64, and split-specific seeds train=42 and validation=43."
+  ],
+  "rai:dataAnnotationProtocol": "No human semantic labels are included in this pretraining dataset. The dataset is intended to be unlabeled audio for unsupervised or self-supervised representation learning.",
+  "rai:annotationsPerItem": "0 human semantic labels per item.",
+  "rai:machineAnnotationTools": [
+    "Hugging Face Datasets Audio feature and parquet conversion tooling were used to package the public dataset."
+  ],
+  "rai:dataUseCases": "[\"Construct represented: segmented 96 kHz passive acoustic recordings of dolphin whistle sequences from the monitored site, with timing, recording-year, duration, and hydrophone metadata.\", \"Validated use case: self-supervised or unsupervised pretraining of audio representation models for dolphin and bioacoustic research, where the objective is to learn acoustic representations rather than supervised labels.\", \"Validated use case: development and validation of feature extractors that may later be fine-tuned on labeled OpenWhistle downstream tasks.\", \"Validated use case: dataset inspection through the review-sample configuration for reviewers or auditors who should not need to download the full dataset.\", \"Supported exploratory use case: analysis of vocal sequences, temporal drift, recording-year coverage, and hydrophone coverage in the released pretraining corpus.\", \"Not validated use case: supervised claims about individual identity, species, behavior, welfare state, conservation status, or deployment performance in other sites or recording conditions without additional labels and local validation.\"]",
+  "rai:dataLimitations": "[\"The dataset is unlabeled and should not be used directly as a supervised benchmark for whistle detection, individual identification, species classification, or behavior classification.\", \"The dataset coverage is uneven across years and hydrophone channels: most examples are from 2021 and 2023, and channel_0 is overrepresented.\", \"The dataset comes from a semi-natural site with a stable pod of known individuals, so model performance may not generalize to fully wild populations, other dolphin species, other locations, recording devices, or acoustic environments without validation.\", \"The review-sample configuration is for inspection only. It preserves split names and approximate structure, but is not a substitute for the full dataset for model development or reporting.\", \"Temporal metadata is provided, but per-example behavioral context, exact animal identity, environmental conditions, and video context are not part of the public columns.\", \"The dataset preserves overlapping vocalizations and natural background noise. This ecological realism is useful for pretraining but can make downstream interpretation difficult.\"]",
+  "rai:dataBiases": "[\"Recording-year distribution is imbalanced: 2021 and 2023 dominate the full dataset relative to 2019, 2020, and 2024.\", \"Hydrophone distribution is imbalanced: channel_0 accounts for most examples, channel_1 is smaller, and channel_2 is rare.\", \"Because the dataset is selected from available recordings and preprocessing decisions, it may overrepresent periods of human presence, periods when dolphins remained near the monitored area, acoustic contexts with higher recording quality, or sequences easier to segment.\", \"The dataset does not encode class labels, so downstream class balance must be assessed separately on labeled fine-tuning or evaluation datasets.\"]",
+  "rai:personalSensitiveInformation": "[\"No human subjects are involved in the dataset. Public columns contain audio, timing fields, year, duration, and hydrophone channel.\", \"The data consists of animal acoustic recordings collected by underwater hydrophones. Users should nevertheless treat incidental non-target sounds as possible in raw acoustic recordings and avoid attempts to identify people, vessels, locations, or sensitive field-site details from the data.\", \"The public pretraining columns do not expose per-example dolphin identities or family relationships.\"]",
+  "rai:dataSocialImpact": "The dataset can support open research on passive acoustic monitoring, dolphin bioacoustics, animal communication, and conservation-oriented acoustic tools. The data collection was non-invasive, did not interfere with dolphin behavior, used passive fixed hydrophones, and involved no human subjects. Misuse risks are limited by the data modality, but include overstating ecological, behavioral, or conservation conclusions from an unlabeled and unevenly sampled pretraining corpus, or deploying models trained on it in new field conditions without local validation.",
+  "rai:hasSyntheticData": false,
+  "rai:dataReleaseMaintenancePlan": "The dataset is hosted publicly on Hugging Face under OpenWhistleNeurIPS26/OpenWhistle-Pretraining and released under CC-BY 4.0. The Hugging Face dataset includes a deterministic review-sample configuration for reviewer inspection of datasets larger than 4 GB. OpenWhistle is part of an ongoing data collection effort, and future releases may expand the audio, extracted whistles, labels, and temporal coverage.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "urn:openwhistle:source-recordings:anonymous-neurips-2026",
+      "prov:label": "Passive acoustic monitoring source recordings for OpenWhistle",
+      "sc:license": "https://creativecommons.org/licenses/by/4.0/",
+      "prov:wasAttributedTo": {
+        "@id": "anonymous_openwhistle_authors",
+        "prov:label": "Anonymous OpenWhistle authors"
+      }
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "Passive acoustic data collection",
+      "sc:description": "Underwater audio was collected passively using fixed and hidden hydrophones. The collection did not interfere with dolphin behavior, did not train or constrain animals, and involved no human subjects.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "anonymous_openwhistle_authors",
+          "prov:label": "Anonymous OpenWhistle authors",
+          "sc:description": "Anonymous research team responsible for the OpenWhistle data release and documentation during double-blind review."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q4929239"
+      },
+      "prov:label": "No semantic annotation for the pretraining corpus",
+      "sc:description": "This pretraining dataset intentionally contains no human semantic labels. There is no labeling schema, annotator instruction set, quality-control adjudication, or inter-annotator agreement score for the released examples. Users requiring supervised evaluation should use the labeled downstream OpenWhistle datasets instead.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "anonymous_openwhistle_authors",
+          "prov:label": "Anonymous OpenWhistle authors",
+          "sc:description": "Anonymous research team responsible for documenting the unlabeled nature of the pretraining release during double-blind review."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q5227332"
+      },
+      "prov:label": "Audio segmentation and dataset packaging",
+      "sc:description": "Raw passive acoustic recordings were segmented into whistle sequences, represented as 96 kHz mono audio examples, and packaged as Hugging Face parquet datasets with timing, duration, year, and hydrophone metadata. The full default configuration contains train and validation splits, while the review-sample configuration is a deterministic shuffled subset for inspection.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "hugging_face_datasets",
+          "prov:label": "Hugging Face Datasets",
+          "sc:description": "Software tooling used to package and publish the dataset with Audio features and parquet conversion."
+        },
+        {
+          "@type": "prov:Agent",
+          "@id": "anonymous_openwhistle_authors",
+          "prov:label": "Anonymous OpenWhistle authors",
+          "sc:description": "Anonymous research team responsible for preprocessing, curation, and release during double-blind review."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": {
+        "@id": "https://www.wikidata.org/wiki/Q3306762"
+      },
+      "prov:label": "Review sample generation",
+      "sc:description": "A deterministic review-sample configuration was generated from the public pretraining dataset to allow reviewers to inspect representative examples without downloading the full dataset. Sampling used base seed 42, shuffle buffer size 64, and split-specific seeds train=42 and validation=43.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:Agent",
+          "@id": "anonymous_openwhistle_authors",
+          "prov:label": "Anonymous OpenWhistle authors",
+          "sc:description": "Anonymous research team responsible for creating the reviewer inspection subset."
+        }
+      ]
+    }
+  ]
+}