Martin Blass, Pejman Mowlaee, W. Bastiaan Kleijn, "Single-Channel
Speech Enhancement Using Double Spectrum", in Proc. Interspeech 2016, 1740-1744. (DOI: 10.21437/Interspeech.2016-234)

- Audio samples -

Below, we present some audio samples demonstrating the impact of the proposed double spectrum speech enhancement versus benchmark methods. The results are shown for the fully blind scenario of male and female utterance corrupted in babble noise. For comparison, results of MMSE-STSA and Modulation Spectral Subtraction (ModSpecSub) methods are shown. Here we present the results of the instrumental predictors for perceived quality and speech intelligibility using PESQ and STOI, respectively:

Female speech: ''Her purse was full of useless trash'' in babble noise SNR = 5 (dB):

Delta-PESQ: Proposed = 0.37, MMSE-STSA = 0.17, ModSpecSub = 0.08

Delta-STOI: Proposed = -0.04, MMSE-STSA = 0.00, ModSpecSub = -0.05

Male speech: ''Her purse was full of useless trash'' in babble noise SNR = 5 (dB):

Delta-PESQ: Proposed = 0.43, MMSE-STSA = 0.22, ModSpecSub = 0.17

Delta-STOI: Proposed = 0.01, MMSE-STSA = -0.02, ModSpecSub = -0.05