Voice Activity Detection and Speech Enhancement in Aviation Noise
Yuan Yao · Audio Engineering · 2006
By analyzing the characteristic of the power spectrum of aviation noise, short-time energy in frequency domain is used to detect voice activity. Transition frames are added based on forward decision and backward decision. What’s more, the transition frames are divided into different groups in detail based on energy. With a further improved method of spectral subtraction, speech frames, transition frames, and noise frames are processed separately to get rid of music noise. Experimental results show that the improved method can offer better performances of the speech under strong aviation noise, both SNR and articulation are improved dramatically.