Frequency domain noise suppression approaches in mobile telephone systems

Frequency-domain noise suppression systems with a single microphone are studied for adverse mobile noise environments. A novel noise suppression algorithm based on a modified maximum likelihood estimate is developed. The algorithm takes into account not only minimum voice distortion but also subjective criteria for the noise naturalness. A speech detector based on the temporal signal-to-noise ratio (SNR) information of every subband of a bandpass filter bank is proposed. The speech detector is proven to be very effective and robust in a rapidly changing noisy background. The proposed noise suppression method in mobile telephone systems demonstrates much better subjective and objective test results than the existing methods. A real-time test shows that 10 dB noise reduction and 9 dB SNR improvements have been achieved even in the most adverse mobile situation.<>