Noise-driven Temporal Trajectory Filtering of Spectral Parameters for Robust Speech Recognition

Spectral parameter filtering such as RASTA [5], RASTA-like bandpass filtering and high-pass filtering operates on temporal dynamics of spectral parameters, and has been effective method to reduce channel distortions. Temporal derivatives (delta and delta-delta coefficients, that have proved as robust representation) and spectral mean normalization [4] are also equivalent to filtering that reject lower modulation frequencies from spectral parameters. Spectral normalization and parameter filtering assume that channel distortions are linear and additive noise is negligible [2]. As the additive noise and channel distortions are additive in different domains, they cannot be simultaneously suppressed by these methods [2]. In this paper, we extend the filtering technique of temporal trajectories to handle additive noise. The paper investigates the possibility of applying ‘spectral subtraction’ [3] filter to time-trajectories of spectral parameters and the problems that may arise.