README.md

metadata

pretty_name: EchoEval
license: apache-2.0
task_categories:
  - audio-classification
  - text-generation
language:
  - en
tags:
  - spoken-empathy
  - speech
  - emotion
  - reasoning
  - benchmark
configs:
  - config_name: normal
    data_files:
      - split: test
        path: normal/test-*
  - config_name: implicit
    data_files:
      - split: test
        path: implicit/test-*
  - config_name: very_high_intense
    data_files:
      - split: test
        path: very_high_intense/test-*
  - config_name: satire_or_self_mockery
    data_files:
      - split: test
        path: satire_or_self_mockery/test-*
  - config_name: shift
    data_files:
      - split: test
        path: shift/test-*

EchoEval

EchoEval is a instance-level spoken empathetic evaluation benchmark. It comprising 1K authentic recordings from 20 professional actors

Load one subset:

from datasets import load_dataset

ds = load_dataset("ddwang2000/EchoEval", "normal", split="test")

Subsets

Subset	Size	Description
normal	220	Explicit, everyday emotional delivery
implicit	220	Emotion is present but understated in the text
very_high_intense	220	High-arousal, strongly expressed emotion
satire_or_self_mockery	220	Ironic, sarcastic, or self-mocking delivery
shift	120	Within-utterance emotion shift (e.g. happy -> sad)

Columns

Column	Type	Notes
sample_id	string	Fresh UUID, not the original internal id
audio	audio	Query utterance
query	string	Transcript / spoken text
category	string	Scene category
sub_category	string	Scene subcategory
audio_emotion	string	Emotion category; shift items use a transition string
polarity	string	positive / negative
intensity	string	normal / high
modality_consistency	string	consistent / conflict
gender	string	female / male
pitch / pitch_level	float / string	Raw value and low / normal / high
energy / energy_level	float / string	Raw value and low / normal / high
speed / speed_level	float / string	Raw value and fast / normal / slow
intonation_contour	string	e.g. rising, falling, rising-falling
stress	list[string]	Stressed words
context_understanding	string	Reasoning gold
latent_need	string	Reasoning gold
core_demand	string	Reasoning gold
strategy_recommended	list[string]	Recommended response strategies
strategy_not_recommended	list[string]	Strategies to avoid
constraints	string	Response constraints

If you find EchoEval useful for your research, we would appreciate an acknowledgment and link to this dataset page. The accompanying paper and its BibTex citation will be added once available.