KMMAU

KMMAU is a Korean audio understanding benchmark constructed from naturally occurring Korean audio. It is part of a Korean speech benchmark suite for evaluating SpeechLMs together with KVoiceBench and KOpenAudioBench.

Dataset Description

KMMAU contains 2,204 Korean audio understanding samples built from KSS, KMSAV, and the Seoul Corpus rather than transferred English items. It evaluates both acoustic and contextual audio understanding capabilities using multiple-choice questions grounded in Korean audio.

KMMAU is constructed from target-language ASR corpora using four construction methods selected by capability:

1. Rule-based generation from speaker metadata for acoustic attributes such as age, gender, and number of speakers.
2. Rule-based generation from transcriptions for lexical questions such as word order and word frequency.
3. LLM-generated questions with human review for semantic questions such as fact extraction and topic summary.
4. Fully manual annotation for holistic capabilities that require listening to the audio, such as general counting and role/profession.

The benchmark contains 646 acoustic samples and 1,558 contextual samples. All KMMAU samples are multiple-choice: gender uses 2 choices, age uses 3 choices, and the remaining capabilities use 4 choices.

Subsets

The dataset consists of 2,204 samples across 9 capabilities:

Config	Category	Samples	Capability
age	Acoustic	276	Speaker age group estimation
gender	Acoustic	270	Speaker gender identification
number_of_speakers	Acoustic	100	Number of speakers detection
word_order	Contextual	1,039	Word order verification
word_frequency_counting	Contextual	180	Word frequency counting in speech
fact_extraction	Contextual	98	Factual information extraction from dialogue
topic_summary	Contextual	100	Topic summarization
role_profession	Contextual	100	Speaker role/profession identification
general_counting	Contextual	41	General counting over audio content

Features

• id (string): Unique sample identifier
• audio (Audio): Korean speech audio
• question (string): Question about the audio
• answer (string): Reference answer
• original_dataset (string): Source dataset name
• original_sample_id (string): Sample identifier in the source dataset
• start_time (float, optional): Audio segment start time in seconds
• end_time (float, optional): Audio segment end time in seconds

Usage

from datasets import load_dataset

# Load a specific subset
ds = load_dataset("anonyemnlpauthor18/KMMAU", "age", split="test")
print(ds[0])

# Load all subsets
subsets = [
    "age", "fact_extraction", "gender", "general_counting",
    "number_of_speakers", "role_profession",
    "topic_summary", "word_frequency_counting", "word_order"
]
for subset in subsets:
    ds = load_dataset("anonyemnlpauthor18/KMMAU", subset, split="test")
    print(f"{subset}: {len(ds)} samples")

Citation

@article{korean_speech_benchmarks2026,
  title={KVoiceBench, KOpenAudioBench, and KMMAU: Agent-Driven Korean Speech Benchmarks for Evaluating SpeechLMs},
  author={Anonymous Authors},
  year={2026}
}

License

This dataset is released under CC BY-NC-SA 4.0.

© 2026 Anonymous Authors