ROG - Ročno označeni govor (Manually annotated speech)

Structure:

ROG consists of 57 manually annotated recordings from the Artur corpus (ROG-Art) and CONLLU files of the same 57 recordings plus additional 287 CONLLU files from the GOS corpus.

ROG
├── DOCS
│   ├── ROG-speakers.tsv
│   └── ROG-speeches.tsv
├── PREBERIME.md
├── README.md
├── ROG-TrainDevTest-split.tsv
├── ROG-Art
│   ├── ROG-Art-TrainDevTest-split.tsv
│   ├── EXB  [57 entries]
│   ├── EXS  [58 entries]
│   ├── TRS  [114 entries]
│   ├── TXT  [114 entries]
│   └── WAV  [57 entries]
└── ROG-SST  [344 entries]

ROG consists of two directories: CONLLU and ROG-Art. CONLLU directory contains 344 conllu files, while ROG-Art is split in subdirectories based on the file types (AVD for audio, TXT for text, ...)

Metadata

Metadata is available for 344 CONLLU files, and separately for the 57 files in the ROG-Art directory. The structure of both variants is the same:

ROG-speakers.tsv

Describes speakers appearing in recordings, and contains columns:

• TEXT-ID: Recording identifier, as used in this corpus
• SOURCE-ID: Recording identifier, as used in their original corpora
• RECORDING-ID: Original recording name
• SUBCORPUS: Subcorpus in GOS2.1 from where the instance was sampled (Gos/Artur-J/Artur-N/Artur-P)
• PRS-ID: Speaker ID
• SEX: Speaker sex
• AGE: Speaker age
• 1LANG: First language of the speaker
• DIALECT: Additional details on the language the speaker is using (described with labels like standardni, pogovorni, narečje, tujejezični vplivi...)
• EDUCATION: Speaker education
• PERM-RESD: Region where the speaker resides
• CHILD-RESD: Region where the speaker lived as a child
• SENTENCES: Number of sentences said by the speaker in the recording
• WORDS: Number of words said by the speaker in the recording.

Minus (-) was used to indicate missing values.

ROG-speeches.tsv

Describes recordings and contains columns:

• TEXT-ID: Recording identifier, as used in this corpus
• SOURCE-ID: Recording identifier, as used in their original corpora
• RECORDING-ID: Original recording name
• SUBCORPUS: Subcorpus in GOS2.1 from where the instance was sampled (Gos/Artur-J/Artur-N/Artur-P)
• TITLE: Description of the recording topic
• DATE: Year and month of the recording
• SOURCE: Recording source, e.g. title of the TV programme from which the speech was sampled
• URL: Recording URL
• TEXT-REGION: Location of the recording
• DOMAIN
• TYPE:
• CHANNEL: Communication channel (e.g. TV / radio / personal contact)
• SENTENCES: Number of sentences
• WORDS: Number of words
• SPK-IDsUTTS: Identifiers of all speakers that appear in the recording

Minus (-) was used to indicate missing values.

ROG-TrainDevTest-split.tsv

Describes composition of train, dev, and test subsets. Contains columns:

• TEXT-ID: Recording identifier
• Split: Subset (Train, Dev, Test)