Nilesh Madhu

Spectrally adaptive loss functions can drastically improve high-frequency speech clarity in streaming applications, addressing a critical gap in current enhancement techniques.

Continuous latent space models can dramatically outperform discrete token approaches in speech enhancement, achieving state-of-the-art results in robustness and quality.

The SFF-based narrowband DoA estimator not only mitigates spatial aliasing but also outperforms leading broadband methods in challenging acoustic environments.