Spaces:
Running
Running
Commit History
-
sync: from Esca01/AInterviewer@c78429518beb4796b649902bde8f73588c887231 Running
Verified signature8d71630 -
sync: from Esca01/AInterviewer@158ecc06673fc78f829855c22238c8595a1d5d63
Verified signatureab61345 -
sync: from Esca01/AInterviewer@87d3050073d1b8d5f0a97de7bb65e9e4b1efe425
AInterviewer CIac695fb
-
fix: retry SER model initialization safely
Verified signature0724421
-
Upload folder using huggingface_hub
Verified signature1f4da5d -
chore(05-02): add huggingface_hub, scikit-learn, joblib to SER space requirements
- Required for SVM artifact download (hf_hub_download + joblib.load) - Required for sklearn Pipeline inference (predict_proba, classes_) - No version pins — Docker resolves latest compatible on build793c460 -
feat(05-02): rewrite SER space with two-stage SVM inference and fallback
- Add svm_pipeline module-level variable with graceful load failure - Add extract_embedding() named function for emotion2vec+ embeddings - Add classify_fallback() named function for emotion2vec+ direct classification - Add conditional /version and /health endpoints based on svm_pipeline state - Use uuid4() temp file naming to prevent concurrent request collisions - Class ordering uses svm_pipeline.classes_ (sklearn alphabetical) not hardcoded list - All 7 tests in tests/ser/test_space.py pass GREEN7f265b3
-
fix(annotate): hide file_name to enforce blind annotation constraint
The filename encodes the raw emotion (e.g. spk_confianza_001.wav), leaking the label before the annotator hears the audio. Removed file_name from the display and added it to _BLIND_COLUMNS. Also: hf_io.py auto-loads .env so scripts work without manual HF_TOKEN export; SER Space switched to wav2vec2-base-superb-er (30s startup vs 30min for emotion2vec_plus_large on free CPU). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a6737c6
-
feat: Migrate Speech Emotion Recognition to a dedicated Hugging Face Space and update client-side integration.
dd6dcad