| language: | |
| - en | |
| pretty_name: "VuiHoc GOPT Gold Features (GOP-80d + Prosody-8d + WavLM-1024d)" | |
| tags: | |
| - goodness-of-pronunciation | |
| - pronunciation-assessment | |
| - wavlm | |
| - gopt | |
| size_categories: | |
| - 1K<n<10K | |
| # VuiHoc GOPT Gold Features (Đã trích xuất đầy đủ Feature) | |
| Dataset GOPT VuiHoc Gold (6,361 câu) đã được trích xuất sẵn toàn bộ Feature trên GPU: | |
| - `feat` [150, 80]: KoelLabs-GOP (log-posterior ratio normalized) | |
| - `occ` [150]: Soft occupancy | |
| - `dur` [150] + `eng` [150, 7]: Prosody 8-d (Viterbi forced alignment + RMSE energy) | |
| - `wavlm` [150, 1024]: WavLM-large Layer 12 pooled per-phone (fp16) | |
| - Nhãn: `phn`, `phone_label`, `word_acc`, `utt_label`, `msdd_type`, `msdd_sub` | |
| ## Load trực tiếp để train | |
| ```python | |
| from datasets import load_dataset | |
| ds = load_dataset("tiennguyenbnbk/gopt-vh-gold-features") | |
| ``` | |