Speech Enhancement in Extremely Noisy Environments Based on Time-Frequency Harmonic Mask with Throat Microphone

We propose a time-frequency harmonic mask (TFHM) utilizing a throat microphone (TM) signal for speech enhancement in extremely noisy environments. The TM captures low-frequency speech signal below 1 kHz without interference from ambient noise. We analyzed the harmonic characteristics of the low-frequency TM signal and replicated this harmonic structure to the high-frequency band up to 4 kHz to generate the TFHM. The TFHM represents the presence of speech as binary values in the time-frequency domain. It is applied to the noisy signal to extract speech accurately. The proposed method demonstrated superior performance over previous approaches through objective and subjective tests under various noise conditions.