Blog(1)

• [INTERSPEECH 2025 Series #5] SPCODEC: Split and Prediction for Neural Speech Codec

  SPCODEC introduces a split-and-prediction scheme to disentangle low/high-frequency components in latent embeddings, enhancing interpretability by aligning spectral features, reducing redundancy, and enabling multi-band encoding within one model.

Research Areas(0)

Publications(0)

News(0)

Others(0)