Name: XiaomiMiMo/SpeechMMLU
Creator: XiaomiMiMo
Published: 2025-09-17 03:37:14
License: 暂无描述

下载链接：

https://hf-mirror.com/datasets/XiaomiMiMo/SpeechMMLU

下载链接

链接失效反馈

官方服务：

资源简介：

--- dataset_info: - config_name: anatomy features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 50709 num_examples: 135 download_size: 23560 dataset_size: 50709 - config_name: clinical_knowledge features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 109162 num_examples: 265 download_size: 45768 dataset_size: 109162 - config_name: college_biology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 72181 num_examples: 144 download_size: 34510 dataset_size: 72181 - config_name: college_medicine features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 106384 num_examples: 172 download_size: 47608 dataset_size: 106384 - config_name: computer_security features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 35575 num_examples: 77 download_size: 18242 dataset_size: 35575 - config_name: econometrics features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 63643 num_examples: 114 download_size: 27150 dataset_size: 63643 - config_name: global_facts features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 32990 num_examples: 99 download_size: 14461 dataset_size: 32990 - config_name: high_school_biology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 164926 num_examples: 309 download_size: 70035 dataset_size: 164926 - config_name: high_school_geography features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 79026 num_examples: 198 download_size: 32392 dataset_size: 79026 - config_name: high_school_government_and_politics features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 112934 num_examples: 193 download_size: 45097 dataset_size: 112934 - config_name: high_school_macroeconomics features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 196297 num_examples: 386 download_size: 65866 dataset_size: 196297 - config_name: high_school_microeconomics features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 124975 num_examples: 238 download_size: 45544 dataset_size: 124975 - config_name: high_school_psychology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 263258 num_examples: 544 download_size: 107651 dataset_size: 263258 - config_name: high_school_us_history features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 2286 num_examples: 2 download_size: 8524 dataset_size: 2286 - config_name: high_school_world_history features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 390450 num_examples: 225 download_size: 198842 dataset_size: 390450 - config_name: human_aging features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 72140 num_examples: 203 download_size: 33278 dataset_size: 72140 - config_name: human_sexuality features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 45020 num_examples: 109 download_size: 23058 dataset_size: 45020 - config_name: international_law features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 74111 num_examples: 121 download_size: 33004 dataset_size: 74111 - config_name: jurisprudence features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 46779 num_examples: 97 download_size: 23961 dataset_size: 46779 - config_name: management features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 34602 num_examples: 103 download_size: 17204 dataset_size: 34602 - config_name: marketing features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 75416 num_examples: 185 download_size: 34985 dataset_size: 75416 - config_name: miscellaneous features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 269620 num_examples: 783 download_size: 116422 dataset_size: 269620 - config_name: moral_disputes features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 161388 num_examples: 342 download_size: 68375 dataset_size: 161388 - config_name: nutrition features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 134635 num_examples: 305 download_size: 61315 dataset_size: 134635 - config_name: philosophy features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 99211 num_examples: 236 download_size: 45912 dataset_size: 99211 - config_name: prehistory features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 123085 num_examples: 289 download_size: 57573 dataset_size: 123085 - config_name: professional_law features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 1477286 num_examples: 1150 download_size: 716159 dataset_size: 1477286 - config_name: professional_psychology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 320537 num_examples: 554 download_size: 137464 dataset_size: 320537 - config_name: public_relations features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 39979 num_examples: 95 download_size: 20782 dataset_size: 39979 - config_name: security_studies features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 235143 num_examples: 238 download_size: 115281 dataset_size: 235143 - config_name: sociology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 94151 num_examples: 201 download_size: 48368 dataset_size: 94151 - config_name: us_foreign_policy features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 45411 num_examples: 100 download_size: 21814 dataset_size: 45411 - config_name: virology features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 61103 num_examples: 166 download_size: 30592 dataset_size: 61103 - config_name: world_religions features: - name: id dtype: string - name: subject dtype: string - name: question_text dtype: string - name: question_audio dtype: string - name: voice_id dtype: string - name: answer dtype: int32 splits: - name: train num_bytes: 53086 num_examples: 171 download_size: 22981 dataset_size: 53086 configs: - config_name: anatomy data_files: - split: train path: anatomy/train-* - config_name: clinical_knowledge data_files: - split: train path: clinical_knowledge/train-* - config_name: college_biology data_files: - split: train path: college_biology/train-* - config_name: college_medicine data_files: - split: train path: college_medicine/train-* - config_name: computer_security data_files: - split: train path: computer_security/train-* - config_name: econometrics data_files: - split: train path: econometrics/train-* - config_name: global_facts data_files: - split: train path: global_facts/train-* - config_name: high_school_biology data_files: - split: train path: high_school_biology/train-* - config_name: high_school_geography data_files: - split: train path: high_school_geography/train-* - config_name: high_school_government_and_politics data_files: - split: train path: high_school_government_and_politics/train-* - config_name: high_school_macroeconomics data_files: - split: train path: high_school_macroeconomics/train-* - config_name: high_school_microeconomics data_files: - split: train path: high_school_microeconomics/train-* - config_name: high_school_psychology data_files: - split: train path: high_school_psychology/train-* - config_name: high_school_us_history data_files: - split: train path: high_school_us_history/train-* - config_name: high_school_world_history data_files: - split: train path: high_school_world_history/train-* - config_name: human_aging data_files: - split: train path: human_aging/train-* - config_name: human_sexuality data_files: - split: train path: human_sexuality/train-* - config_name: international_law data_files: - split: train path: international_law/train-* - config_name: jurisprudence data_files: - split: train path: jurisprudence/train-* - config_name: management data_files: - split: train path: management/train-* - config_name: marketing data_files: - split: train path: marketing/train-* - config_name: miscellaneous data_files: - split: train path: miscellaneous/train-* - config_name: moral_disputes data_files: - split: train path: moral_disputes/train-* - config_name: nutrition data_files: - split: train path: nutrition/train-* - config_name: philosophy data_files: - split: train path: philosophy/train-* - config_name: prehistory data_files: - split: train path: prehistory/train-* - config_name: professional_law data_files: - split: train path: professional_law/train-* - config_name: professional_psychology data_files: - split: train path: professional_psychology/train-* - config_name: public_relations data_files: - split: train path: public_relations/train-* - config_name: security_studies data_files: - split: train path: security_studies/train-* - config_name: sociology data_files: - split: train path: sociology/train-* - config_name: us_foreign_policy data_files: - split: train path: us_foreign_policy/train-* - config_name: virology data_files: - split: train path: virology/train-* - config_name: world_religions data_files: - split: train path: world_religions/train-* --- # Dataset Card for SpeechMMLU ## Dataset Description SpeechMMLU is an evaluation dataset designed to assess the knowledge capabilities of speech-language models. It is built based on the [MMLU](https://huggingface.co/datasets/cais/mmlu) dataset, with entries filtered by subject and length, resulting in a total of 8,549 entries across 34 subjects. We synthesized the questions and answers into speech using commercial TTS with diverse voices. This dataset can be used for evaluating knowledge capabilities of speech-language models in text-to-text, speech-to-text, text-to-speech, and speech-to-speech scenarios. ## Dataset Structure This dataset comprises 34 subjects, including: - `anatomy` - `clinical_knowledge` - `college_biology` - `college_medicine` - `computer_security` - `econometrics` - `global_facts` - `high_school_biology` - `high_school_geography` - `high_school_government_and_politics` - `high_school_macroeconomics` - `high_school_microeconomics` - `high_school_psychology` - `high_school_us_history` - `high_school_world_history` - `human_aging` - `human_sexuality` - `international_law` - `jurisprudence` - `management` - `marketing` - `miscellaneous` - `moral_disputes` - `nutrition` - `philosophy` - `prehistory` - `professional_law` - `professional_psychology` - `public_relations` - `security_studies` - `sociology` - `us_foreign_policy` - `virology` - `world_religions` Each subject contains several entries, and each entry includes the following fields: 1. `id`: A unique identifier. 2. `subject`: The subject category. 3. `question_text`: The text of the question and options. 4. `question_audio`: The corresponding audio for the question and options. 5. `voice_id`: The voice ID used in TTS. 6. `answer`: A numeric value (0, 1, 2, or 3) indicating the correct answer (A, B, C, or D). Additionally, the dataset includes two files: 1. `audio.tar.gz`: A compressed archive containing all related audio files. 2. `few_shot_prompts.json`: Each subject contains 5 example questions for few-shot evaluation. > **Note**: During few-shot evaluation, it is important to select few-shot prompts that use the same voice as the current question. Therefore, we have synthesized few-shot prompts in different voices, and the corresponding files are included in the `audio.tar.gz` archive. ## Contact For any questions or further details about the dataset, please contact us at [mimo@xiaomi.com](mailto:mimo@xiaomi.com).

应用场景：