AudioHate - Audio Hate Speech Detection

# AudioHate - Audio Hate Speech Detection with Explanations
# Based on Guo et al., SIGDIAL 2024
# Paper: https://aclanthology.org/2024.sigdial-1.35/
# Dataset: https://github.com/audio-hate/AudioHateXplain
#
# Task: Classify audio clips for hate speech, identify target groups,
# and note acoustic indicators (tone, emphasis, prosody).
#
# Guidelines:
# - Listen to the full audio clip carefully before annotating
# - Consider both speech content and acoustic features (tone, emphasis)
# - Distinguish between hate speech, offensive language, and neither
# - Identify all applicable target groups when hate/offensive speech is present
# - Rate your confidence in the classification

annotation_task_name: "AudioHate: Explainable Audio Hate Speech Detection"
task_dir: "."

data_files:
  - sample-data.json
item_properties:
  id_key: "id"
  text_key: "audio_url"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  - annotation_type: radio
    name: hate_speech_classification
    description: "Classify this audio clip for hate speech content"
    labels:
      - name: "Hate Speech"
        tooltip: "Speech that attacks or demeans a group based on protected attributes"
        key_value: "h"
      - name: "Offensive"
        tooltip: "Offensive or inappropriate language that is not targeting a specific group"
        key_value: "o"
      - name: "Neither"
        tooltip: "No hate speech or offensive language detected"
        key_value: "n"

  - annotation_type: multiselect
    name: target_groups
    description: "If hate speech or offensive, select ALL targeted groups (leave empty if Neither)"
    labels:
      - name: "Race"
        tooltip: "Targeting based on race or ethnicity"
      - name: "Gender"
        tooltip: "Targeting based on gender or gender identity"
      - name: "Religion"
        tooltip: "Targeting based on religious beliefs or affiliation"
      - name: "Sexuality"
        tooltip: "Targeting based on sexual orientation"
      - name: "Disability"
        tooltip: "Targeting based on physical or mental disability"
      - name: "Nationality"
        tooltip: "Targeting based on national origin or citizenship"
      - name: "Other"
        tooltip: "Targeting based on other group characteristics"

  - annotation_type: multiselect
    name: acoustic_indicators
    description: "Select any notable acoustic indicators in the speech"
    labels:
      - name: "Aggressive tone"
        tooltip: "Speaker uses a hostile or aggressive vocal tone"
      - name: "Mocking tone"
        tooltip: "Speaker uses a sarcastic or mocking vocal pattern"
      - name: "Emphasis on slurs"
        tooltip: "Speaker places special emphasis on derogatory terms"
      - name: "Raised voice"
        tooltip: "Speaker is shouting or speaking loudly"
      - name: "Threatening prosody"
        tooltip: "Speech rhythm and intonation suggest a threat"
      - name: "Sarcastic intonation"
        tooltip: "Speaker uses sarcastic vocal inflection"

  - annotation_type: radio
    name: confidence
    description: "How confident are you in your classification?"
    labels:
      - name: "Very confident"
        tooltip: "Clear and unambiguous classification"
        key_value: "1"
      - name: "Somewhat confident"
        tooltip: "Fairly certain but some ambiguity"
        key_value: "2"
      - name: "Not confident"
        tooltip: "Difficult to determine, significant ambiguity"
        key_value: "3"

audio_display:
  show_waveform: true
  playback_controls: true
  allow_speed_control: true

allow_all_users: true
instances_per_annotator: 100
annotation_per_instance: 3
allow_skip: true
skip_reason_required: false