saeeew/JP-HomophoneBench

1---2language:3- ja4task_categories:5- automatic-speech-recognition6license: other7pretty_name: JP-HomophoneBench8tags:9- japanese10- asr11- homophone12- phoneme13- contextual-asr14- context-biasing15- speech-recognition16configs:17- config_name: homophone818  data_files:19  - split: train20    path: homophone8/train-00000-of-00001.parquet21  - split: validation22    path: homophone8/validation-00000-of-00001.parquet23  - split: test24    path: homophone8/test-00000-of-00001.parquet25- config_name: homophone8-audio26  data_files:27  - split: test28    path: homophone8-audio/test-*29- config_name: homophone8-research30  data_files:31  - split: train32    path: homophone8-research/train-00000-of-00001.parquet33  - split: validation34    path: homophone8-research/validation-00000-of-00001.parquet35  - split: test36    path: homophone8-research/test-00000-of-00001.parquet37dataset_info:38  config_name: homophone8-audio39  features:40  - name: audio_filepath41    dtype: 'null'42  - name: audio_ref43    struct:44    - name: config45      dtype: string46    - name: kind47      dtype: string48    - name: repo_id49      dtype: string50    - name: row_id51      dtype: string52    - name: split53      dtype: string54  - name: candidates55    list: json56  - name: category57    dtype: string58  - name: difficulty59    struct:60    - name: acoustic61      dtype: float6462    - name: context63      dtype: float6464    - name: lexical65      dtype: float6466    - name: phone_distance67      dtype: float6468  - name: group_id69    dtype: string70  - name: id71    dtype: string72  - name: metadata73    dtype: json74  - name: source75    struct:76    - name: audio_redistributed77      dtype: bool78    - name: config79      dtype: string80    - name: dataset81      dtype: string82    - name: license83      dtype: string84    - name: revision85      dtype: string86    - name: source_id87      dtype: string88    - name: split89      dtype: string90    - name: synthetic91      dtype: bool92  - name: split93    dtype: string94  - name: target95    dtype: json96  - name: text97    dtype: string98  - name: benchmark_split99    dtype: string100  - name: audio101    dtype:102      audio:103        sampling_rate: 16000104  - name: audio_origin105    dtype: string106  - name: audio_engine107    dtype: string108  - name: audio_engine_version109    dtype: string110  - name: audio_voice111    dtype: string112  - name: audio_voice_license113    dtype: string114  - name: audio_source_config115    dtype: string116  - name: audio_source_revision117    dtype: string118  - name: audio_sampling_rate119    dtype: int64120  splits:121  - name: test122    num_bytes: 3761077123    num_examples: 71124  download_size: 3748747125  dataset_size: 3761077126---127 128# JP-HomophoneBench129 130A deterministic Japanese ASR benchmark index for separating eight error/disambiguation classes:131 132- `exact_homophone`133- `near_homophone`134- `voicing`135- `long_vowel`136- `geminate`137- `moraic_nasal`138- `pitch_accent`139- `semantic_only`140 141## Important design rule142 143This repository is metadata-first. Source audio is **not redistributed by default**. Each row stores source repository/config/split/row identifiers so audio can be rehydrated under the original source license.144 145`exact_homophone` and `semantic_only` are intentionally separate. A phoneme scorer is not expected to solve `semantic_only` by itself.146 147HF repository: `saeeew/JP-HomophoneBench`148Most recently published configuration: `homophone8`149Publication license policy: `permissive`150 151## Available configurations152 153- `homophone8`154- `homophone8-research`155 156## Published rows in this run157 158### Splits159 160- `train`: 111161- `validation`: 9162- `test`: 11163 164### Categories165 166- `exact_homophone`: 10167- `geminate`: 3168- `long_vowel`: 3169- `moraic_nasal`: 3170- `near_homophone`: 7171- `pitch_accent`: 92172- `semantic_only`: 10173- `voicing`: 3174 175### Upstream sources176 177- `HaitongSUN/prosodic-abx`: 92178- `JP-HomophoneBench/permissive-phonetic-core`: 19179- `manual-cc0-example`: 20180 181## Licensing and provenance182 183The combined benchmark uses `license: other` because rows retain their upstream license. Inspect `source.license` before redistribution or commercial use.184 185## Loading186 187```python188from datasets import load_dataset189ds = load_dataset("saeeew/JP-HomophoneBench", "homophone8")190print(ds)191```192 193## Build summary194 195```json196{197  "augmentation": "permissive-core8-v0.1",198  "categories": {199    "exact_homophone": 178,200    "geminate": 9,201    "long_vowel": 18,202    "moraic_nasal": 5,203    "near_homophone": 210,204    "pitch_accent": 92,205    "semantic_only": 178,206    "voicing": 17207  },208  "core8_missing": [],209  "licenses": {210    "cc-by-4.0": 92,211    "cc-by-nc-4.0": 576,212    "cc0-1.0": 39213  },214  "records": 707,215  "sha256": {216    "all.jsonl": "897940216f5f1130b0954e3381f0ec37d998575791f2b170e3add3c5b664eeb3",217    "test.jsonl": "a657d1accff04dd7c1c629a4ef1fe8232b11ed2425a962187feee5fff6aebb43",218    "train.jsonl": "93a3fa069fefada5564a2f550a9ce98579fe611da4c6afcb43b5e8e755f5f18d",219    "validation.jsonl": "12e5d99945e844d8be53aea4f2864a912b56bb907458291ed698d25cd60f564c"220  },221  "sources": {222    "HaitongSUN/prosodic-abx": 92,223    "IDEMITSU/hoiku-yougo-stt-ja": 576,224    "JP-HomophoneBench/permissive-phonetic-core": 19,225    "manual-cc0-example": 20226  },227  "splits": {228    "test": 161,229    "train": 475,230    "validation": 71231  }232}233```234