JSON Data File Structure

Each JSON data file contains three objects: corpus, system and dataset.

corpus

corpus provides basic corpus information:

{"corpus": {
    "identifier": String,
    "license": String,
    "documentation_link": String,
    "download_link": String,
    "cost": Boolean,
    "incomplete_utterances_removed_%": Float,
    "missing_corpus_data": Dictionary,
    "original_reference_segmentation": String,
    "original_audio_segmentation": String
}

system

system contains identifier (either the system full configuration or its ID) as well as the information on the model and language code that were applied for the particular transcription:

"system": {
    "identifier": String,
    "model": String,
    "language_code": String
}

dataset

dataset provides detailed information about the utterance set including metadata and basic statistics as well as the utterances list itself:

"dataset": {
    "language": String,
    "speaking_style": String,
    "num_utterances": Integer,
    "total_duration": Float,
    "average_audio_duration": Float,
    "num_speaker_noise_references": Integer,
    "%_overlapping_utterances": Float,
    "max_num_overlapping_utterances": Integer,
    "num_utterances_containing_non_lexical_sounds": Integer,
    "num_utterances_containing_only_non_lexical_sounds": Integer,
    "average_speaking_rate": Float,
    "recording_devices": String,
    "acoustic_environments": String,
    "overlapping_speech": String,
    "utterances_per_gender": Dictionary,
    "utterances_per_dialect": Dictionary,
    "utterances_per_accent": Dictionary,
    "num_speakers": Integer,
    "max_num_utterances_per_speaker": Integer,
    "min_num_utterances_per_speaker": Integer,
    "avg_num_utterances_per_speaker": Float,
    "utterances": List
}

For more information on the utterance structure go to Utterance Structure.