7 days ago
-
Model card: correct the voice-share interval of Small
-
Model card: disclose that the sealed recordings were in an earlier student's unlabelled distillation audio
-
Model card: report the sealed speakers with listener-checked labels; drop the held-out set
-
Name the sizes nano, tiny, mini and small
-
Nested Accent Detective: one network, four sizes, optional text
10 days ago
-
Add fast mode (wav2vec 2.0 path only) next to the accurate mode
-
Accent Detective v3: dual-teacher student (wav2vec 2.0 + Whisper), nested judges, syllable-identity normalisation
11 days ago
-
Accent Detective v2: student distilled on more varied audio (AISHELL-1, THCHS-30); evaluated on held-out speakers
-
Wording: Taiwan vs not Taiwan
-
safetensors checkpoint; inference code only
-
Docstring wording
-
Model card: training data wording
-
Model card: BibTeX citation
-
Model card: shorter, no semicolons
-
Accent Detective: compact Taiwan-accent judge (G17)
-
initial commit
