Phase Estimation for Improved Single-Channel Source Separation Based on Time-Frequency Masking

- Audio samples -




- Phase enhancement (PE) combined with Time-Frequency Masking (TFM) -


Estimated Binary Mask

Male Utter + Speech Shaped Noise (SSN) at SNR = 0 (dB)

Uttering: “The small red neon lamp went out”




Estimated Ratio Mask

Male Utter + cafeteria noise at SNR = 0 (dB)

Uttering: “The fan whirled its round blades softly”