formosan-asr-taroko / evaluation_results.json
hunterschep's picture
Publish current best Formosan ASR checkpoint for Taroko
253f94b verified
Raw
History Blame Contribute Delete
5.51 kB
{
"aggregate": {
"cer_normalized_micro": 0.025800467225505015,
"character_errors": 751,
"corpora": 4,
"reference_characters": 29108,
"reference_words": 6991,
"utterances": 962,
"wer_normalized_micro": 0.060506365326848804,
"word_errors": 423
},
"by_corpus": [
{
"cer_normalized": "0.026968",
"cer_raw": "0.026968",
"char_deletions": "96",
"char_deletions_raw": "96",
"char_insertions": "97",
"char_insertions_raw": "97",
"char_substitutions": "130",
"char_substitutions_raw": "130",
"eval_corpus": "ILRDF",
"eval_language": "taroko",
"eval_utterances": "474",
"reference_characters": "11977",
"reference_characters_raw": "11977",
"reference_words": "2881",
"reference_words_raw": "2881",
"run_id": "whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"wer_normalized": "0.083652",
"wer_raw": "0.083652",
"word_deletions": "12",
"word_deletions_raw": "12",
"word_insertions": "17",
"word_insertions_raw": "17",
"word_substitutions": "212",
"word_substitutions_raw": "212"
},
{
"cer_normalized": "0.046866",
"cer_raw": "0.046866",
"char_deletions": "38",
"char_deletions_raw": "38",
"char_insertions": "26",
"char_insertions_raw": "26",
"char_substitutions": "22",
"char_substitutions_raw": "22",
"eval_corpus": "ePark1",
"eval_language": "taroko",
"eval_utterances": "125",
"reference_characters": "1835",
"reference_characters_raw": "1835",
"reference_words": "420",
"reference_words_raw": "420",
"run_id": "whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"wer_normalized": "0.090476",
"wer_raw": "0.090476",
"word_deletions": "4",
"word_deletions_raw": "4",
"word_insertions": "1",
"word_insertions_raw": "1",
"word_substitutions": "33",
"word_substitutions_raw": "33"
},
{
"cer_normalized": "0.073411",
"cer_raw": "0.073411",
"char_deletions": "2",
"char_deletions_raw": "2",
"char_insertions": "93",
"char_insertions_raw": "93",
"char_substitutions": "69",
"char_substitutions_raw": "69",
"eval_corpus": "ePark2",
"eval_language": "taroko",
"eval_utterances": "58",
"reference_characters": "2234",
"reference_characters_raw": "2234",
"reference_words": "531",
"reference_words_raw": "531",
"run_id": "whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"wer_normalized": "0.043315",
"wer_raw": "0.043315",
"word_deletions": "12",
"word_deletions_raw": "12",
"word_insertions": "2",
"word_insertions_raw": "2",
"word_substitutions": "9",
"word_substitutions_raw": "9"
},
{
"cer_normalized": "0.013627",
"cer_raw": "0.013627",
"char_deletions": "62",
"char_deletions_raw": "62",
"char_insertions": "39",
"char_insertions_raw": "39",
"char_substitutions": "77",
"char_substitutions_raw": "77",
"eval_corpus": "ePark3",
"eval_language": "taroko",
"eval_utterances": "305",
"reference_characters": "13062",
"reference_characters_raw": "13062",
"reference_words": "3159",
"reference_words_raw": "3159",
"run_id": "whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"wer_normalized": "0.038303",
"wer_raw": "0.038303",
"word_deletions": "9",
"word_deletions_raw": "9",
"word_insertions": "9",
"word_insertions_raw": "9",
"word_substitutions": "103",
"word_substitutions_raw": "103"
}
],
"condition": "pan_formosan_to_language",
"dataset_version": "dataset_v1",
"iso_639_3": "trv",
"language": "Taroko",
"language_id": "taroko",
"model_family": "whisper_small",
"normalization_version": "formosan_safe_v1",
"readiness": {
"actual_train_rows": 7650,
"best_selection_cer": 0.026846856509768723,
"best_selection_wer": 0.061902180140190874,
"best_step": 1722,
"condition": "pan_formosan_to_language",
"dominant_hypothesis_rate": 0.005197505197505198,
"estimated_effective_epochs": 3.0002614379084966,
"failures": 0,
"generated_at": "2026-06-25T21:34:57+00:00",
"intended_train_rows": 7650,
"manifest_dir": "/projects/prudlab/tacl_formosan_asr/experiments/manifests/gate5_whisper_final_v1_with_pan_checkpoint/whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"nonempty_hypothesis_rate": 1.0,
"prediction_rows": 962,
"result_dir": "/scratch/scheppat/projects/tacl_formosan_asr/runs_whisper_final_v1/whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"result_label": "final",
"run_id": "whisper_small_pan_formosan_to_language_train-taroko_eval-taroko-all-corpora_dataset_v1_split_v2_formosan_safe_v1_seed13",
"status": "pass",
"steps": 2869,
"test_rows": 962,
"train_batch_size": 8,
"warnings": 0
},
"seed": 13,
"split_version": "split_v2"
}