Figure 12From: On pre-image iterations for speech enhancementComparison of the spectrograms after application of (a) PID and (b) the MMSE log-spectral amplitude estimator on the example utterance corrupted by AWGN at 15 dB SNR. The MMSE log-spectral amplitude estimator removes noise near speech components more efficiently and performs less attenuation on low energy speech components.Back to article page