README.md

metadata

dataset_info:
  features:
    - name: qid
      dtype: int64
    - name: instruction
      dtype: string
    - name: answer
      dtype: string
    - name: source_dataset
      dtype: string
    - name: task_metric
      dtype: string
    - name: attribute_type
      dtype: string
    - name: attribute_value
      dtype: string
    - name: audio1
      dtype: audio
    - name: audio2
      dtype: audio
    - name: audio3
      dtype: audio
    - name: audio4
      dtype: audio
    - name: audio5
      dtype: audio
    - name: audio1_duration
      dtype: float64
    - name: audio2_duration
      dtype: float64
    - name: audio3_duration
      dtype: float64
    - name: audio4_duration
      dtype: float64
    - name: audio5_duration
      dtype: float64
    - name: audio1_sps
      dtype: float64
    - name: audio2_sps
      dtype: float64
    - name: audio3_sps
      dtype: float64
    - name: audio4_sps
      dtype: float64
    - name: audio5_sps
      dtype: float64
    - name: audio1_text
      dtype: string
    - name: audio2_text
      dtype: string
    - name: audio3_text
      dtype: string
    - name: audio4_text
      dtype: string
    - name: audio5_text
      dtype: string
  splits:
    - name: test_4
      num_bytes: 22143838
      num_examples: 50
    - name: test_3
      num_bytes: 16703789
      num_examples: 50
    - name: test_2
      num_bytes: 11270268
      num_examples: 50
    - name: test
      num_bytes: 27337007
      num_examples: 50
  download_size: 77471486
  dataset_size: 77454902
configs:
  - config_name: default
    data_files:
      - split: test_4
        path: data/test_4-*
      - split: test_3
        path: data/test_3-*
      - split: test_2
        path: data/test_2-*
      - split: test
        path: data/test-*

Multi-Audio-Grounding: Max-Attribute Selection (L2-ARCTIC + CREMA-D)

Task: Given 5 audio options (A–E) and an instruction, select the option that matches a target attribute and is either the longest (duration) or the fastest (speaking rate / sps) among the matching options.

Composition (50 questions)

By dataset

• CREMA-D: 36
• L2-ARCTIC: 14

By metric

• Fastest (sps): 25
• Longest (duration): 25

By attribute type

• emotion: 22
• gender: 28

Breakdown (dataset × metric × attribute_type)

dataset	metric	attribute_type	count
CREMA-D	Fastest (sps)	emotion	11
CREMA-D	Fastest (sps)	gender	7
CREMA-D	Longest (duration)	emotion	11
CREMA-D	Longest (duration)	gender	7
L2-ARCTIC	Fastest (sps)	gender	7
L2-ARCTIC	Longest (duration)	gender	7