datasone / BS-GSA

BS-GSA: Speech Emotion Recognition via Blend-Sample Empowered Global & Spectral Attentive Paradigm

Replicating our results

• Download our pre-extracted feature file from Google Drive (personal share).
  • Warning: clicking the Google Drive share link will immediately reveal the author identity
  • The file is split to zip files, so the partitioned_features file will need to be decompressed
• Create a experiment directory for storing log, data and models
• Run train_test.sh in the experiment directory
  • ../train_test.sh <FEATURE_FILE>

---

• For more usage, see full code for model structure, data processing and training.