Feature Extraction
Transformers
Safetensors
PyTorch
Russian
accentik
russian
accentuation
word-stress
custom-code
custom_code
Instructions to use TeraSpace/accentik with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use TeraSpace/accentik with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("feature-extraction", model="TeraSpace/accentik", trust_remote_code=True)# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("TeraSpace/accentik", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
Download evaluation.json from TeraSpace/accentik: direct link, hf CLI and curl.
- Browser
- Download file 6.36 kB
-
https://huggingface.co/TeraSpace/accentik/resolve/main/evaluation.json
- Command line
-
hf download hf://TeraSpace/accentik/evaluation.json
-
curl -L -o evaluation.json https://huggingface.co/TeraSpace/accentik/resolve/main/evaluation.json
6.36 kB
| { | |
| "release": "audio-v3", | |
| "reference_release": "experimental-context-v2", | |
| "reference_weights_sha256": "8e0be4ecb637a35c4f152e2cb887bca8e95ab1f1b17daa4f62a1706331f7aa3e", | |
| "source_checkpoint_sha256": "f54918eb8146ee85fa429f1dffe30cf8c2adce3132b057fe2d9adc35b1f311ba", | |
| "stress_checkpoint_sha256": "5342342f6469de3f985dbc0799a3083f2ed8f979488f920812fb6a911125bd54", | |
| "stress_weights_bitwise_preserved": true, | |
| "input_yo_hidden_for_paired_comparison": true, | |
| "stress_checks": { | |
| "RHD_manual": { | |
| "label_kind": "manual contextual reference", | |
| "targets": 7676, | |
| "context_v2_correct": 7482, | |
| "audio_v3_correct": 7484, | |
| "context_v2_score": 0.9747264200104221, | |
| "audio_v3_score": 0.9749869723814487, | |
| "context_v2_macro": 0.9637496548122184, | |
| "audio_v3_macro": 0.9679068553070598 | |
| }, | |
| "external_russtress": { | |
| "label_kind": "public reference, previously reused during development", | |
| "targets": 11089, | |
| "context_v2_correct": 10435, | |
| "audio_v3_correct": 10483, | |
| "context_v2_score": 0.9410226350437371, | |
| "audio_v3_score": 0.9453512489854811, | |
| "context_v2_macro": 0.9361600330721678, | |
| "audio_v3_macro": 0.9425269193674681 | |
| }, | |
| "TeraTTS/stress_dataset_sft_proza": { | |
| "label_kind": "supplied labels", | |
| "targets": 21234, | |
| "context_v2_correct": 19900, | |
| "audio_v3_correct": 19913, | |
| "context_v2_score": 0.9371762268060657, | |
| "audio_v3_score": 0.9377884524818687, | |
| "context_v2_macro": 0.8681086799998965, | |
| "audio_v3_macro": 0.8710729274781607 | |
| }, | |
| "TeraTTS/open_accent": { | |
| "label_kind": "supplied labels", | |
| "targets": 46994, | |
| "context_v2_correct": 43577, | |
| "audio_v3_correct": 44045, | |
| "context_v2_score": 0.9272885900327701, | |
| "audio_v3_score": 0.9372473081670001, | |
| "context_v2_macro": 0.8973879348784138, | |
| "audio_v3_macro": 0.9089840690228318 | |
| }, | |
| "TeraTTS/stress_dataset_sft_poetry": { | |
| "label_kind": "supplied labels", | |
| "targets": 5882, | |
| "context_v2_correct": 4981, | |
| "audio_v3_correct": 4995, | |
| "context_v2_score": 0.846820809248555, | |
| "audio_v3_score": 0.8492009520571234, | |
| "context_v2_macro": 0.811748154183098, | |
| "audio_v3_macro": 0.8215229111497144 | |
| }, | |
| "supplied_dialogue": { | |
| "label_kind": "RUAccent pseudo-label agreement", | |
| "targets": 42965, | |
| "context_v2_correct": 42483, | |
| "audio_v3_correct": 42685, | |
| "context_v2_score": 0.9887815663912487, | |
| "audio_v3_score": 0.9934830676131735, | |
| "context_v2_macro": 0.9561116028270102, | |
| "audio_v3_macro": 0.9954060892784302 | |
| }, | |
| "agent_authored": { | |
| "label_kind": "authored development diagnostics; not representative", | |
| "targets": 12, | |
| "context_v2_correct": 12, | |
| "audio_v3_correct": 7, | |
| "context_v2_score": 1.0, | |
| "audio_v3_score": 0.5833333333333334, | |
| "context_v2_macro": 1.0, | |
| "audio_v3_macro": 0.5833333333333334 | |
| }, | |
| "audio/portal_ru": { | |
| "label_kind": "filtered acoustic teacher agreement", | |
| "targets": 58, | |
| "context_v2_correct": 58, | |
| "audio_v3_correct": 57, | |
| "context_v2_score": 1.0, | |
| "audio_v3_score": 0.9827586206896551, | |
| "context_v2_macro": 1.0, | |
| "audio_v3_macro": 0.9827586206896551 | |
| } | |
| }, | |
| "manual_context_macro": { | |
| "context_v2": 0.9637496548122184, | |
| "audio_v3": 0.9679068553070598 | |
| }, | |
| "acoustic_teacher_agreement": { | |
| "context_v2": 0.9921076879932695, | |
| "audio_v3": 0.9965546252153359 | |
| }, | |
| "yo": { | |
| "experimental": true, | |
| "enabled_by_default": false, | |
| "validation_selected_threshold": 0.8, | |
| "known_manual_targets": { | |
| "tp": 736, | |
| "fp": 34, | |
| "fn": 49, | |
| "tn": 2818, | |
| "precision": 0.9558441558441558, | |
| "recall": 0.9375796178343949, | |
| "f1": 0.9466237942122186, | |
| "accuracy": 0.9771789936761067 | |
| }, | |
| "authored_target_diagnostic": { | |
| "correct": 39, | |
| "total": 45, | |
| "positive_correct": 24, | |
| "positive_total": 27, | |
| "negative_correct": 15, | |
| "negative_total": 18 | |
| }, | |
| "authored_whole_text_diagnostic": { | |
| "sentences": 45, | |
| "whole_sentence_correct": 36, | |
| "tp": 28, | |
| "fp": 6, | |
| "fn": 6, | |
| "precision": 0.8235294117647058, | |
| "recall": 0.8235294117647058 | |
| }, | |
| "diagnostic_representative": false, | |
| "passed_default_restoration_gate": false | |
| }, | |
| "data": { | |
| "train_contexts": 472192, | |
| "validation_contexts": 54434, | |
| "test_contexts": 67048, | |
| "usable_audio_texts": 49996, | |
| "filtered_acoustic_word_targets": 323918, | |
| "unknown_labels_masked": true, | |
| "source_texts_redistributed": false | |
| }, | |
| "label_provenance": { | |
| "clarified_on": "2026-10-28", | |
| "supplied_dialogue": { | |
| "stress_label_generator": "RUAccent", | |
| "provenance_evidence": "User confirmed that the Russian dialogue data was stressed by RUAccent.", | |
| "metric_kind": "RUAccent pseudo-label agreement", | |
| "independent_human_reference": false, | |
| "interpretation": "An increase measures closer agreement with RUAccent, including its errors. Held-out contexts do not make automatic labels an independent correctness reference.", | |
| "numeric_scores_changed": false | |
| }, | |
| "audio_ru_dialogs": { | |
| "metric_kind": "Filtered TeraAnaly acoustic-teacher agreement", | |
| "teacher_training_caveat": "TeraAnaly used RUAccent-derived training labels.", | |
| "independent_human_reference": false | |
| }, | |
| "reference_comparisons": { | |
| "sources": [ | |
| "RHD_manual", | |
| "external_russtress" | |
| ], | |
| "interpretation": "Report separately from pseudo-label agreement. Public references have been reused during development and do not establish a score on the private benchmark." | |
| } | |
| }, | |
| "limitations": [ | |
| "Dialogue labels were generated by RUAccent; agreement includes teacher errors.", | |
| "Acoustic scores measure agreement with TeraAnaly, itself trained with RUAccent-derived labels.", | |
| "Public reference tests were reused during development; historical model/data overlap is unknown.", | |
| "Authored diagnostics are development regression checks, not a representative human benchmark.", | |
| "Optional ё placement can invent or miss ё and can disagree with predicted stress.", | |
| "No private benchmark examples or expected answers were used; no claim of beating RUAccent." | |
| ] | |
| } | |